For this project, either use your own Kaggle API and proceed with the code or download the dataset on kaggle website
Tutorial to create Kaggle API key in Python: https://stackoverflow.com/questions/49310470/using-kaggle-datasets-in-google-colab
Link to manually download the dataset : https://www.kaggle.com/fedesoriano/heart-failure-prediction/download
Estimated runtime : about 2 hours
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
import numpy as np
#! pip install scikit-optimize
import skopt
from scipy import stats
import scipy
#!pip install hypopt
import hypopt
import zipfile
#!pip install -q kaggle
import kaggle
#Run this if you use Kaggle API key
#!kaggle datasets download -d fedesoriano/heart-failure-prediction
#Run this to unzip the dataset and create a folder containing the csv dataset
#with zipfile.ZipFile('heart-failure-prediction.zip', 'r') as zip_ref:
# zip_ref.extractall('heart-failure-prediction')
# zip_ref.close()
original_df=pd.read_csv('heart-failure-prediction/heart.csv')
original_df
| Age | Sex | ChestPainType | RestingBP | Cholesterol | FastingBS | RestingECG | MaxHR | ExerciseAngina | Oldpeak | ST_Slope | HeartDisease | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | 40 | M | ATA | 140 | 289 | 0 | Normal | 172 | N | 0.0 | Up | 0 |
| 1 | 49 | F | NAP | 160 | 180 | 0 | Normal | 156 | N | 1.0 | Flat | 1 |
| 2 | 37 | M | ATA | 130 | 283 | 0 | ST | 98 | N | 0.0 | Up | 0 |
| 3 | 48 | F | ASY | 138 | 214 | 0 | Normal | 108 | Y | 1.5 | Flat | 1 |
| 4 | 54 | M | NAP | 150 | 195 | 0 | Normal | 122 | N | 0.0 | Up | 0 |
| ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... | ... |
| 913 | 45 | M | TA | 110 | 264 | 0 | Normal | 132 | N | 1.2 | Flat | 1 |
| 914 | 68 | M | ASY | 144 | 193 | 1 | Normal | 141 | N | 3.4 | Flat | 1 |
| 915 | 57 | M | ASY | 130 | 131 | 0 | Normal | 115 | Y | 1.2 | Flat | 1 |
| 916 | 57 | F | ATA | 130 | 236 | 0 | LVH | 174 | N | 0.0 | Flat | 1 |
| 917 | 38 | M | NAP | 138 | 175 | 0 | Normal | 173 | N | 0.0 | Up | 0 |
918 rows × 12 columns
Information available in Kaggle data main page
original_df.isna().info()
<class 'pandas.core.frame.DataFrame'> RangeIndex: 918 entries, 0 to 917 Data columns (total 12 columns): # Column Non-Null Count Dtype --- ------ -------------- ----- 0 Age 918 non-null bool 1 Sex 918 non-null bool 2 ChestPainType 918 non-null bool 3 RestingBP 918 non-null bool 4 Cholesterol 918 non-null bool 5 FastingBS 918 non-null bool 6 RestingECG 918 non-null bool 7 MaxHR 918 non-null bool 8 ExerciseAngina 918 non-null bool 9 Oldpeak 918 non-null bool 10 ST_Slope 918 non-null bool 11 HeartDisease 918 non-null bool dtypes: bool(12) memory usage: 10.9 KB
No missing value 'per say' in the dataset
original_df.dtypes.value_counts()
int64 6 object 5 float64 1 dtype: int64
There are 7 numerical data and 5 categorical data. However, looking at the data type, some of the numerical value are encoded categorical values. For instance, FastingBS is a categorical variable that is one for diabete and 0 for normal. The HeartDisease is the same. So the true number of cat/num data is 5/7
cat_index=np.asarray(original_df.columns[original_df.dtypes==object])
cat_index=np.insert(cat_index,2,'FastingBS')
num_index=np.asarray(original_df.columns[original_df.dtypes=='int64'])
num_index=np.delete(num_index,(3,5))
num_index=np.insert(num_index,4,'Oldpeak')
print('Categorical predictors :'+str(cat_index)+'\nNumerical predictors'+str(num_index))
Categorical predictors :['Sex' 'ChestPainType' 'FastingBS' 'RestingECG' 'ExerciseAngina' 'ST_Slope'] Numerical predictors['Age' 'RestingBP' 'Cholesterol' 'MaxHR' 'Oldpeak']
There are 11 predictors, and the response is a binary response, with 1 being Heart Disease and 0 is normal
Now, to study the different predictors associations, it is needed to understand the type of the variables used.
Indeed, some variables here are categorical dichotomous (binary), some are categorical non-dichotomous and some are numerical.
The correlation is only meaningful for 2 numerical variables together. To study the association between two non-numerical variable, the Pearson Correlation isn't relevant, and other coefficients must be used
To solve this issue, the proposed solution is:
with $N$ the sample size,$k$ the lesser number of categories and $\chi^{2}$ the chi-square value. </li>
</ul>
To measure the association between a categorical and a non categorical data, there are two options:
#create a plot with two variables depending on their type
def pairwise_plot(x1,x2,ax,type_):
if type_==1:
#only numerical_variables -> scatterplot
ax.scatter(x1,x2)
ax.set_xlabel(x1.name)
ax.set_ylabel(x2.name)
elif type_==2:
#only categorical values -> stacked barplot
pd.crosstab(x1,x2,normalize='index').plot(kind='bar',stacked=True,cmap='Set2',ax=ax)
elif type_==3:
#cat and num values -> for each category a curve is drawn counting the number of appearance
#of the categorical variable in the y axis, and the numerical value on the x axis
if x2.dtype==object or x2.name=='FastingBS' or x2.name=='HeartDisease':
x_temp=x1
x1=x2
x2=x_temp
else:
pass
cross=pd.crosstab(x1,x2).iloc[1,:].index
for i in range(0,pd.crosstab(x1,x2).index.size):
cat_count=pd.crosstab(x1,x2).iloc[i,:]
ax.plot(cross,cat_count,label=pd.crosstab(x1,x2).index[i])
ax.legend()
ax.set_xlabel(x2.name)
ax.set_ylabel(x1.name+' count')
return ax
def correlation_calculation(data,data_type=[],plot_=False):
if plot_==True:
fig,ax=plt.subplots(len(data_type),len(data_type),figsize=(100,100))
else:
pass
#get type of the data, important for chosing the right correlation calculation
alpha=0.05
n_features=data.columns.value_counts().sum()
if len(data_type)==0:
data_type=np.asarray(data.dtypes,str)
else:
data_type=np.asarray(data_type)
correlation_matrix=np.empty((data_type.size,data_type.size))
for i in range(0,data_type.size):
x1=data.iloc[:,i]
for j in range(0,data_type.size):
x2=data.iloc[:,j]
if data_type[i]!='object' and data_type[j]!='object':
#'Pearson correlation for two numerical
type_=1
x1=np.asarray(x1)
x2=np.asarray(x2)
x_mean=np.array([x1.mean(),x2.mean()])
s=np.array([np.sqrt(np.sum(np.square(x1-x_mean[0]))),
np.sqrt(np.sum(np.square(x2-x_mean[1])))])
r=(np.sum((x1-x_mean[0])*(x2-x_mean[1]))/(s[0]*s[1]))
elif data_type[i]=='object' and data_type[j]=='object':
#'Cramer v for two categorical
type_=2
cont_table=pd.crosstab(x1,x2)
stat,p,dof,expected=stats.chi2_contingency(cont_table)
if p>alpha:
r=0
else:
r=stat/((min(cont_table.shape)-1)*x1.size)
elif np.unique(x1).size==2 or np.unique(x2).size==2:
#if one of the two data is binary point and there are one cat and one num -> point biserial
type_=3
if np.unique(x1).size==2 :
x_bin=np.where(x1==np.unique(x1)[0],0,1)
pb=stats.pointbiserialr(x_bin,x2)
elif np.unique(x2).size==2:
x_bin=np.where(x2==np.unique(x2)[0],0,1)
pb=stats.pointbiserialr(x_bin,x1)
if pb[1]>alpha:
r=0
else:
r=pb[0]
else:
#if cat+num without binary variable --> cramer's v
type_=3
cont_table=pd.crosstab(x1,x2)
stat,p,dof,expected=stats.chi2_contingency(cont_table)
if p>alpha:
r=0
else:
r=stat/((min(cont_table.shape)-1)*x1.size)
correlation_matrix[i,j]=r
if plot_==True:
pairwise_plot(data.iloc[:,i],data.iloc[:,j],ax=ax[i,j],type_=type_)
df_corr=pd.DataFrame(correlation_matrix,columns=data.columns,index=data.columns)
return df_corr
data_type=['int64','object','object','int64','int64','object','object','int64','object','float64','object','object']
corr_mat=correlation_calculation(original_df,data_type=data_type,plot_=True)
corr_mat
| Age | Sex | ChestPainType | RestingBP | Cholesterol | FastingBS | RestingECG | MaxHR | ExerciseAngina | Oldpeak | ST_Slope | HeartDisease | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Age | 1.000000 | 0.000000 | 0.083348 | 0.254399 | -0.095282 | 0.198039 | 0.078584 | -0.382045 | 0.215793 | 0.258612 | 0.084695 | 0.282039 |
| Sex | 0.000000 | 0.993450 | 0.040173 | 0.000000 | -0.200092 | 0.013669 | 0.000000 | -0.189186 | 0.035322 | 0.105734 | 0.023764 | 0.091661 |
| ChestPainType | 0.083348 | 0.040173 | 1.000000 | 0.000000 | 0.000000 | 0.025820 | 0.010545 | 0.175170 | 0.195287 | 0.112718 | 0.085449 | 0.292012 |
| RestingBP | 0.254399 | 0.000000 | 0.000000 | 1.000000 | 0.100893 | 0.070193 | 0.100112 | -0.112135 | 0.155101 | 0.164803 | 0.000000 | 0.107589 |
| Cholesterol | -0.095282 | -0.200092 | 0.000000 | 0.100893 | 1.000000 | -0.260974 | 0.000000 | 0.235792 | 0.000000 | 0.050148 | 0.000000 | -0.232741 |
| FastingBS | 0.198039 | 0.013669 | 0.025820 | 0.070193 | -0.260974 | 0.993916 | 0.016583 | -0.131438 | 0.000000 | 0.000000 | 0.030903 | 0.070066 |
| RestingECG | 0.078584 | 0.000000 | 0.010545 | 0.100112 | 0.000000 | 0.016583 | 1.000000 | 0.000000 | 0.011569 | 0.094782 | 0.000000 | 0.011908 |
| MaxHR | -0.382045 | -0.189186 | 0.175170 | -0.112135 | 0.235792 | -0.131438 | 0.000000 | 1.000000 | -0.370425 | -0.160691 | 0.201924 | -0.400421 |
| ExerciseAngina | 0.215793 | 0.035322 | 0.195287 | 0.155101 | 0.000000 | 0.000000 | 0.011569 | -0.370425 | 0.995482 | 0.408752 | 0.208528 | 0.242113 |
| Oldpeak | 0.258612 | 0.105734 | 0.112718 | 0.164803 | 0.050148 | 0.000000 | 0.094782 | -0.160691 | 0.408752 | 1.000000 | 0.266253 | 0.403951 |
| ST_Slope | 0.084695 | 0.023764 | 0.085449 | 0.000000 | 0.000000 | 0.030903 | 0.000000 | 0.201924 | 0.208528 | 0.266253 | 1.000000 | 0.387711 |
| HeartDisease | 0.282039 | 0.091661 | 0.292012 | 0.107589 | -0.232741 | 0.070066 | 0.011908 | -0.400421 | 0.242113 | 0.403951 | 0.387711 | 0.995597 |
This pairwise comparison plot is made for every single x1 and x2 combination.
The strategy was the following:
plt.figure(figsize=(14,8))
ax=sns.heatmap(corr_mat,cmap='bwr')
ax.set_title('Dataset Correlation Heat Map',fontsize=18)
plt.show()
#fig = ax.get_figure()
#fig.savefig("Dataset_Correlation_Heat_Map.png")
There are no strong correlation between each predictors. The maximum correlation coefficient is 0.4, which can be considered a medium-strength correlation.
It is important to notice that the cholesterol has a negative correlation with the Heart Disease risk, which is quiet unintuitive. </br> A further analysis of the predictors compared to the response is presented herebelow
#Function to quickly plot a numerical data against the heart disease prediction.
def plot_overall(feature_name):
fig,ax=plt.subplots(2,figsize=(20,35))
x1=pd.crosstab(original_df.iloc[:,11],original_df[feature_name]).iloc[1,:].index
heartdisease_count=pd.crosstab(original_df.iloc[:,11],original_df[feature_name]).iloc[1,:]
ax[0].set_title('Frequency of Heart Disease depending on the '+feature_name+' of patients',fontsize=18)
ax[0].set_xlabel(feature_name)
ax[0].set_ylabel('Count')
ax[0].plot(x1,heartdisease_count,label='HeartDisease')
ax[0].plot(x1,original_df[feature_name].value_counts().sort_index(),label='Repartition of '+feature_name)
ax[0].grid()
ax[0].legend()
#plot a less biased representation of the numerical feature, using percentage of the total
#sampling population instead of counting the number of case.
ax[1].plot(heartdisease_count/original_df[feature_name].value_counts().sort_index()*100)
ax[1].grid()
ax[1].set_xlabel(feature_name)
ax[1].set_ylabel('Percentage of the sample population having a heartdisease')
ax[1].set_title('Repartition of heart disease in percentage over the '+feature_name+' sample',fontsize=18)
plt.show()
plot_overall('Age')
The graphic on the top represents the number of heart disease case per age, compared to the distribution of age over the entire sample population.
The number of cases by age follow the age distribution of the entire sample population.
It is needed to unbias the data from the original distribution.
To do so, the percentage of the sample population at each age is used as the response. To illustrate, if 4 people in the sample population have 74 years old, and 3 get an heart disease at this age, the response will be 75% for the age 74
Comparing the total sampling population to the affected population over age, it seems that there is in fact a direct correlation between Age and risk of Heart Disease (which is expected, according to current medical knowledge).
The older one get, the more likely a heart disease will occur. The plot of occurence of heart disease over age is skewed due to the distribution of the sample population age, which isn't uniform as it is shown herebelow. It seems to follow a gaussian distribution with a mean around 50
sns.set(rc={'figure.figsize':(14,10)})
sns.displot(x=original_df.iloc[:,0],kde=True,height=16,legend=False)
plt.suptitle("Distribution of the age over the sample population",
fontsize=24)
plt.show()
sns.histplot(data=original_df,x='Sex',hue='HeartDisease',multiple='dodge').set_title('Number of heart disease per sex')
plt.show()
pd.crosstab(original_df.iloc[:,1],original_df.iloc[:,11])
| HeartDisease | 0 | 1 |
|---|---|---|
| Sex | ||
| F | 143 | 50 |
| M | 267 | 458 |
print('Percentage of women with heart disease : '+str(round(50/(50+143)*100,2))+'%')
print('Percentage of men with heart disease : '+str(round(458/(267+458)*100,2))+'%')
Percentage of women with heart disease : 25.91% Percentage of men with heart disease : 63.17%
For the sex of the patient, 63% of men got a heart disease, and only about 26% of women. Here the difference is pretty significant, men are more likely to suffer an heart disease than woman.
sns.histplot(original_df,x='ChestPainType',hue='HeartDisease',multiple='dodge').set_title('Chest Pain Type impact on heart disease risk ')
plt.show()
ax=sns.catplot(data=original_df,x='ChestPainType',y='HeartDisease',kind='bar',ci=None,height=6)
ax.fig.subplots_adjust(top=0.92)
ax.fig.suptitle('Percentage of Heart Disease per ChestPainType')
plt.show()
These two plots shows the type of chest pain count for both population, and the percentage of each type of pain with heart disease. Two things can be said from these plot.
sns.displot(x=original_df.iloc[:,3],kde=True,height=16,legend=False)
plt.suptitle("Distribution of the Resting blood pressure over the sample population",
fontsize=24)
plt.show()
The Resting Blood Pressure distribution seems to be gaussian centered on 130. There is one outlier to drop, with a resting bp of 0, which basically would mean dead.
plot_overall('RestingBP')
For the frequency plot, it doesn't seems to have once again a particular trend. It shows that the number of heart disease increases with the number of people in the one category of age. </br> </br>Then the repartition of heart disease depending on the RestingBP is a quiet noisy graph. Moreover, one value makes no sense (0), because the blood pressure is expected to be a positive number. It seems though that the percentage increases with the RestingBP once this odd value being taken out. Applying a filter to denoise the data clearly confirm that assumption.
The chosen filter is the Savitzky-Golay filter, because of its simplicity of use. It fits a n order polymomial for each i, using k points for the prediction. The new value is then $\hat{f}(i)$.
In the example herebelow, it fits a linear model using 21 points for prediction.
from scipy.signal import savgol_filter
x1=(pd.crosstab(original_df.iloc[:,11],original_df.iloc[:,3]).iloc[1,:].index)[1:]
heartdisease_count=pd.crosstab(original_df.iloc[:,11],original_df.iloc[:,3]).iloc[1,:]
k=21
n=1
yhat = savgol_filter((heartdisease_count/original_df.iloc[:,3].value_counts().sort_index()*100).iloc[1:], k, n)
plt.plot((heartdisease_count/original_df.iloc[:,3].value_counts().sort_index()*100).iloc[1:],label='Original Data')
plt.title('Repartition of Heart Disease over the total sample population')
plt.plot(x1,yhat,label='Smoothed Data')
plt.xlabel('Resting Blood Pressure')
plt.ylabel('Percentage of the sample population affected')
plt.legend()
plt.show()
sns.displot(x=original_df.iloc[:,4],kde=True,height=16,legend=False)
plt.suptitle("Distribution of the Cholesterol over the sample population",
fontsize=24)
plt.show()
The cholesterol level is a bit tricky. The 0 represents a big proportion of the sample population, so it would be better to keep it in the dataset
</br></br>
The very high value are sever case of cholesterol, these are possible values.
First of all, it is important to see the responses associated to this 0
pd.crosstab(original_df.iloc[:,4][original_df.iloc[:,4]==0],original_df.iloc[:,11][original_df.iloc[:,4]==0],normalize='index')
| HeartDisease | 0 | 1 |
|---|---|---|
| Cholesterol | ||
| 0 | 0.116279 | 0.883721 |
Most of the 0 are associated with a heart disease (over 88%). Knowing that, it is very likely that these 0 come from a dataset where cholesterol wasn't reported. (Data comes from several different dataset).
It is obvious that these runs cannot be used as it is, because the model will be biased by these missing values.
Because, all of the other features for this data are relevant, it would be better to find a way to keep these values.
Some simple solutions are proposed in a first stage:
For the first option, the strategy is to set a threshold splitting high cholesterol and low cholesterol into two categories
np_cho=original_df.iloc[:,4].copy()
for i in range(0,np_cho.size):
if np_cho[i]<200:
np_cho[i]=0
else:
np_cho[i]=1
cho_unique,cho_count=np.unique(np_cho,return_counts=True)
plt.bar(cho_unique,height=cho_count)
plt.xlabel('Cholesterol presence')
plt.xticks([0,1])
plt.ylabel('Count')
plt.title('Repartition of Cholesterol presence')
plt.show()
pd.crosstab(np_cho,original_df.iloc[:,11],normalize='index').plot(kind='bar',stacked=True,cmap='Set2')
plt.title('Heart Disease presence compared to Cholesterol presence')
plt.show()
pb=stats.pointbiserialr(np_cho,original_df.iloc[:,11])
pb
PointbiserialrResult(correlation=-0.15208541911533593, pvalue=3.6808518091207524e-06)
Both value are binary, so the Point biserial correlation is a better indicator of a possible correlation between the variables.
When treated as categorical, the cholesterol level has a weak correlation with the risk of heart disease.
On this sample, it is important to notice that about 37% of people without cholesterol are
healthy, and about 50% for people with cholesterol.
These numbers are against the current medical knowledge. A high cholesterol is known to induce cardiac problems
Another strategy is to deal with the cholesterol value as numerical value. But to do so, it is needed to 'guess' the possible values for the missing cholesterol. Knowing that about 90% of the missing cholesterol population had an heart disease, one guess would be that the average cholesterol value for this population was quiet high.
Then it is possible to generate a list of cholesterol value following a gaussian distribution and centered on a high cholesterol value.
cholesterol_x=original_df.iloc[:,4].copy()
number=cholesterol_x[cholesterol_x==0].value_counts()
mu=220
sd=20
new_distri=np.random.normal(mu,sd,number)
ax=sns.displot(x=new_distri,kde=True,height=16,legend=False)
ax.fig.subplots_adjust(top=0.95)
ax.fig.suptitle('Distribution of the randomly generated data')
plt.show()
np_co=np.asarray(cholesterol_x)
n=0
for i in range(0,np_co.size):
if np_co[i]==0:
np_co[i]=new_distri[n]
n=n+1
else:
pass
sns.displot(x=np_co,kde=True,height=16,legend=False)
plt.suptitle("Distribution of the Cholesterol over the sample population with randomly generated values",
fontsize=24)
plt.subplots_adjust(top=0.95)
plt.show()
cross_value=pd.crosstab(np_co,original_df.iloc[:,11])
for i in range (0,cross_value.shape[0]):
cross_value.iloc[i,0],cross_value.iloc[i,1]=cross_value.iloc[i,:]/(cross_value.iloc[i,0]+cross_value.iloc[i,1])
plt.plot(cross_value.index,cross_value[1])
plt.title('Heart disease risk depending on the cholesterol level')
plt.xlabel('Cholesterol level')
plt.ylabel('Percentage of the sample population affected')
plt.show()
yhat_1 = savgol_filter(cross_value.iloc[:,1],21, 2)
plt.plot(cross_value.index,yhat_1)
plt.title('Heart disease risk depending on the cholesterol level with smoothed data')
plt.xlabel('Cholesterol level')
plt.ylabel('Percentage of the sample population affected')
plt.show()
There are no clear evidence of any impact of the cholesterol level over the risk of getting an heart disease using a gaussian distribution to replace the missing value.
pb=stats.pointbiserialr(np_co,original_df.iloc[:,11])
pb
PointbiserialrResult(correlation=0.03395507500171286, pvalue=0.3040989898337783)
Moreover the point biserial result is very low, meaning that there is a very weak correlation between this new data and the risk of heart disease.
It is needed to replace this missing values, thus an entire part of this project will be dedicated to the imputation of a new cholesterol value.
pd.crosstab(original_df.iloc[:,5],original_df.iloc[:,11],normalize='index').plot(kind='bar',stacked=True,cmap='Set2')
plt.title('Repartition of Fasting Blood Sugar')
plt.show()
ax=sns.catplot(data=original_df,x='FastingBS',y='HeartDisease',kind='bar',ci=None,height=8)
ax.fig.subplots_adjust(top=0.92)
ax.fig.suptitle('Percentage of Heart Disease per Fasting Blood Sugar Presence')
plt.show()
pd.crosstab(original_df.iloc[:,5],original_df.iloc[:,11])
| HeartDisease | 0 | 1 |
|---|---|---|
| FastingBS | ||
| 0 | 366 | 338 |
| 1 | 44 | 170 |
print('Percentage of the sample population without diabete affected by heart disease : '+str(round(338/(366+338)*100,2))+'%')
print('Percentage of the smaple population with diabete affected by heart disease : '+str(round(170/(170+44)*100,2))+'%')
Percentage of the sample population without diabete affected by heart disease : 48.01% Percentage of the smaple population with diabete affected by heart disease : 79.44%
Fasting Blood Sugar is an indicator of Diabete. According to the data, and medical knowledge, diabete increases the odds of Heart Disease. In fact, without diabete, the odds are almost 50/50 of getting an heart disease. However, the presence of diabete makes this value rise to 80%.
pd.crosstab(original_df.iloc[:,6],original_df.iloc[:,11],normalize='index').plot(kind='bar',stacked=True,cmap='Set2')
plt.title('Repartition of Resting ECG')
plt.show()
ax=sns.catplot(data=original_df,x='RestingECG',y='HeartDisease',kind='bar',ci=None,height=6)
ax.fig.subplots_adjust(top=0.92)
ax.fig.suptitle('Percentage of Heart Disease depending on the Resting ECG')
plt.show()
pd.crosstab(original_df.iloc[:,6],original_df.iloc[:,11])
| HeartDisease | 0 | 1 |
|---|---|---|
| RestingECG | ||
| LVH | 82 | 106 |
| Normal | 267 | 285 |
| ST | 61 | 117 |
print('Percentage of the sample population showing a left ventricular hypertrophy affected : '+str(round(106/(106+82)*100,2))+'%')
print('Percentage of the sample population showing a ST-T wave abnormality affected : '+str(round(117/(117+61)*100,2))+'%')
print('Percentage of the sample population showing a normal ECG affected : '+str(round(285/(285+267)*100,2))+'%')
Percentage of the sample population showing a left ventricular hypertrophy affected : 56.38% Percentage of the sample population showing a ST-T wave abnormality affected : 65.73% Percentage of the sample population showing a normal ECG affected : 51.63%
For the resting ECG, it is very mitigated. One would have guess that a normal ECG means a lower risk of Heart Disease, but this is slightly the case. An abnormal ECG increases the odds of Heart Disease (ie. ST).
sns.displot(x=original_df.iloc[:,7],kde=True,height=16,legend=False)
plt.suptitle("Distribution of the maximum heart rate over the sample population",
fontsize=24)
plt.show()
There are no clear outliers for the MaxHR variable. Once again, the repartition of the values could be approximated by a gaussian distribution centered on 130-140.
plot_overall('MaxHR')
The Maximum heart rate is one of the most correlated feature with the response according to the correlation matrix, with a correlation of -0.4 . Even though the count is skewed by the sample distribution, which is not uniform, the repartition in percentage shows a clear decreasing trend between the percentage of population having a heart disease an the maximum HR.
pd.crosstab(original_df.iloc[:,8],original_df.iloc[:,11],normalize='index').plot(kind='bar',stacked=True,cmap='Set2')
plt.title('Repartition of Exercice Angina')
plt.show()
ax=sns.catplot(data=original_df,x='ExerciseAngina',y='HeartDisease',kind='bar',ci=None,height=6)
ax.fig.subplots_adjust(top=0.92)
ax.fig.suptitle('Percentage of Heart Disease depending on Exercice Angina ')
plt.show()
There is an obvious correlation between Heart Disease and Exercise Angina (chest pain due to exercise), as 35% of population without Exercise Angina had an Heart Disease, but around 85% for those with Exercise Angina
plot_overall('Oldpeak')
Oldpeak is a measurement of the deviation over a flat line of an ECG segment. This is measured as a depression, and negative values should become positive.
original_df.Oldpeak=abs(original_df.Oldpeak)
plot_overall('Oldpeak')
The Oldpeak is clearly correlated with heartdisease, as expected. The higher it gets, the higher the odds of heart disease gets. And it makes sense, as this value depicts how weak is the heart
Before diving into the analysis of this feature, it is better to understand its meaning. On an ECG, after an heartbeat, the signal values sent to the heart decreases, becoming negative after a peak. The ST slope is then the trend of the section between the S and the start of T (in the picture below). Usually it tends to have a slight upward trend. However, in some cases, mostly related with myocardial infection, it can have a a flat or downslopping shape.

pd.crosstab(original_df.iloc[:,10],original_df.iloc[:,11],normalize='index').plot(kind='bar',stacked=True,cmap='Set2')
plt.show()
ax=sns.catplot(data=original_df,x='ST_Slope',y='HeartDisease',kind='bar',ci=None,height=6)
ax.fig.subplots_adjust(top=0.92)
ax.fig.suptitle('Percentage of Heart Disease depending on the ST Slope type ')
plt.show()
pd.crosstab(original_df.iloc[:,10],original_df.iloc[:,11])
| HeartDisease | 0 | 1 |
|---|---|---|
| ST_Slope | ||
| Down | 14 | 49 |
| Flat | 79 | 381 |
| Up | 317 | 78 |
print('Percentage of the sample population showing a downward trend affected : '+str(round(49/(49+14)*100,2))+'%')
print('Percentage of the sample population showing a flat trend affected : '+str(round(381/(381+79)*100,2))+'%')
print('Percentage of the sample population showing an upward trend affected : '+str(round(78/(78+317)*100,2))+'%')
Percentage of the sample population showing a downward trend affected : 77.78% Percentage of the sample population showing a flat trend affected : 82.83% Percentage of the sample population showing an upward trend affected : 19.75%
ST Slope is one of the highest correlated feature with heart disease. Indeed, as the graphs shows, an heart disease is very likely to happen if the slope is flat or downslopping. The risks are increased by 400% if the slope is not going upward
Now that every single feature has been analyzed a first time, some conclusion can be drawn, and data can be transformed toward a more meaningful shape.
Two actions need to be taken:
Removing the outlier is a fairly simple operation:
original_df[original_df.RestingBP==0]
| Age | Sex | ChestPainType | RestingBP | Cholesterol | FastingBS | RestingECG | MaxHR | ExerciseAngina | Oldpeak | ST_Slope | HeartDisease | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 449 | 55 | M | NAP | 0 | 0 | 0 | Normal | 155 | N | 1.5 | Flat | 1 |
original_df.drop(index=449,axis=0,inplace=True)
original_df[original_df.RestingBP==0]
| Age | Sex | ChestPainType | RestingBP | Cholesterol | FastingBS | RestingECG | MaxHR | ExerciseAngina | Oldpeak | ST_Slope | HeartDisease |
|---|
original_df.Cholesterol[original_df.Cholesterol==0].value_counts()
0 171 Name: Cholesterol, dtype: int64
There are 171 missing values.
In order to keep the 170 runs with the cholesterol missing values, it is necessary to find a way to replace these 0s.
The strategy here is to find a model predicting the cholesterol value that have a fairly good prediction rate
The metric used to compare these imputing methods is chosen to be the accuracy of the classifiers
First, it is needed to transform the categorical variable, one hot encoding them.
#insert dummy variables to get rid of categorical variables
knn_df=original_df.copy()
dummy_1=pd.get_dummies(knn_df.iloc[:,1])
dummy_2=pd.get_dummies(knn_df.iloc[:,2])
dummy_6=pd.get_dummies(knn_df.iloc[:,6])
dummy_8=pd.get_dummies(knn_df.iloc[:,8])
dummy_10=pd.get_dummies(knn_df.iloc[:,10])
knn_df.drop(['Sex','ChestPainType','RestingECG','ExerciseAngina','ST_Slope'],axis=1,inplace=True)
knn_df=knn_df.join(dummy_1).join(dummy_2).join(dummy_6).join(dummy_8).join(dummy_10)
#getting the different data organized
# main code from sklearn
from sklearn.preprocessing import StandardScaler,MinMaxScaler,RobustScaler,PowerTransformer,normalize
from sklearn.ensemble import RandomForestClassifier
from sklearn.experimental import enable_iterative_imputer
from sklearn.impute import SimpleImputer, KNNImputer, IterativeImputer
from sklearn.model_selection import cross_val_score, GridSearchCV, RandomizedSearchCV,train_test_split
from sklearn.pipeline import make_pipeline
from sklearn.linear_model import BayesianRidge, ElasticNet
from sklearn.tree import DecisionTreeRegressor
from sklearn.ensemble import ExtraTreesRegressor
from sklearn.neighbors import KNeighborsRegressor
import xgboost as xgb
from sklearn.metrics import roc_auc_score
#classifier used to compare the different imputers, randomly chosen
clf = RandomForestClassifier()
def get_train_test_val(X_full,y_full,test_ratio):
#function to permorm two train test split in a row, in order to get a train, test and validation set
X_full_train, X_test, y_full_train, y_test = train_test_split(
X_full, y_full, test_size=test_ratio, stratify=y_full
)
X_train, X_val, y_train, y_val = train_test_split(
X_full_train, y_full_train, test_size=test_ratio, stratify=y_full_train
)
return X_train,y_train,X_test,y_test,X_val,y_val
#
def get_clean_score(X_clean,y_clean,clf=clf):
#get score if no imputer is used. All the patients with missing cholesterol are dropped out
clean_scores = cross_val_score(
clf, scaler.fit_transform(X_clean), y_clean, scoring='accuracy', cv=10
)
return clean_scores.mean(), clean_scores.std()
def get_scores_for_imputer(imputer,X_full, y_full,clf=clf):
#perform a crossvalidation on imputed data using clf
estimator = make_pipeline(imputer,clf)
impute_scores = cross_val_score(
estimator, scaler.fit_transform(X_full), y_full, scoring='accuracy', cv=10
)
return impute_scores
def get_simple_impute_score(X_full, y_full,strategy):
imputer = SimpleImputer(
missing_values=np.nan, strategy=strategy, fill_value=0, add_indicator=True
)
simple_impute_scores = get_scores_for_imputer(imputer, X_full, y_full)
return simple_impute_scores.mean(), simple_impute_scores.std()
def get_iterative_impute_score(X_full, y_full, strategy):
imputer = IterativeImputer(
missing_values=np.nan, estimator=strategy, add_indicator=True,max_iter=20
)
iterative_impute_scores = get_scores_for_imputer(imputer, X_full, y_full)
return iterative_impute_scores.mean(), iterative_impute_scores.std()
def plot_imputers(x_labels,acc_mean,acc_std):
fig, ax = plt.subplots(figsize=(13, 10))
ax.barh(x_labels, acc_mean, xerr=acc_std, align='center', alpha=0.5, ecolor='black', capsize=10)
ax.set_ylabel('Imputing Technique')
ax.set_xlabel('Accuracy score of the prediction')
ax.set_title('Accuracy score of prediction of different Imputing techniques')
ax.yaxis.grid(True)
plt.xlim([0.75,1])
The data is split this way:
scaler=StandardScaler()
X_full=knn_df.drop('HeartDisease',axis=1).copy()
X_full.Cholesterol.replace(0,np.nan,inplace=True)
X_clean=X_full[X_full.Cholesterol.isnull()==False]
X_missing=X_full[X_full.Cholesterol.isnull()==True].copy()
y_full=knn_df.HeartDisease
y_clean=knn_df[knn_df.Cholesterol!=0].HeartDisease
y_missing=knn_df[knn_df.Cholesterol==0].HeartDisease
The use of the sklearn.imput class allows a quick lookout on different simple and elaborated imputing techniques.
The simple imputers are:
With iterative imputers, a model is fitted to predict the missing values, the different models are:
simple_strategy=['constant','median','mean']
iterative_strategy=[BayesianRidge(),
DecisionTreeRegressor(max_features="sqrt"),
ExtraTreesRegressor(n_estimators=15),
KNeighborsRegressor(n_neighbors=15)]
x_labels=['Clean Data','Simple fill w/ 0','Simple Median','Simple Mean','Iterative w/ BayesianRidge',
'Iterative w/ Decision Tree','Iterative w/ ExtraTrees','Iterative with KN Reg',
'Cholesterol data dropped']
The base code has been built over this piece of code to rank different imputers.
acc_mean = np.zeros(9)
acc_std=np.zeros(9)
i=0
acc_mean[0], acc_std[0] = get_clean_score(scaler.fit_transform(X_clean), y_clean)
i+=1
for strat in simple_strategy:
acc_mean[i], acc_std[i] = get_simple_impute_score(X_full, y_full,strategy=strat)
i+=1
for strat in iterative_strategy:
acc_mean[i], acc_std[i] = get_iterative_impute_score(X_full, y_full,strategy=strat)
i+=1
chol_drop=cross_val_score(clf, scaler.fit_transform(X_full.drop('Cholesterol',axis=1))
, y_full, scoring="accuracy", cv=10)
acc_mean[i], acc_std[i]=chol_drop.mean(),chol_drop.std()
pd.DataFrame(np.c_[acc_mean,acc_std],
columns=['Prediction accuracy','Prediction Standard Deviation'],
index=x_labels)
C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\impute\_iterative.py:699: ConvergenceWarning: [IterativeImputer] Early stopping criterion not reached. warnings.warn( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\impute\_iterative.py:699: ConvergenceWarning: [IterativeImputer] Early stopping criterion not reached. warnings.warn( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\impute\_iterative.py:699: ConvergenceWarning: [IterativeImputer] Early stopping criterion not reached. warnings.warn( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\impute\_iterative.py:699: ConvergenceWarning: [IterativeImputer] Early stopping criterion not reached. warnings.warn( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\impute\_iterative.py:699: ConvergenceWarning: [IterativeImputer] Early stopping criterion not reached. warnings.warn( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\impute\_iterative.py:699: ConvergenceWarning: [IterativeImputer] Early stopping criterion not reached. warnings.warn( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\impute\_iterative.py:699: ConvergenceWarning: [IterativeImputer] Early stopping criterion not reached. warnings.warn( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\impute\_iterative.py:699: ConvergenceWarning: [IterativeImputer] Early stopping criterion not reached. warnings.warn( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\impute\_iterative.py:699: ConvergenceWarning: [IterativeImputer] Early stopping criterion not reached. warnings.warn( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\impute\_iterative.py:699: ConvergenceWarning: [IterativeImputer] Early stopping criterion not reached. warnings.warn( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\impute\_iterative.py:699: ConvergenceWarning: [IterativeImputer] Early stopping criterion not reached. warnings.warn( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\impute\_iterative.py:699: ConvergenceWarning: [IterativeImputer] Early stopping criterion not reached. warnings.warn( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\impute\_iterative.py:699: ConvergenceWarning: [IterativeImputer] Early stopping criterion not reached. warnings.warn( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\impute\_iterative.py:699: ConvergenceWarning: [IterativeImputer] Early stopping criterion not reached. warnings.warn( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\impute\_iterative.py:699: ConvergenceWarning: [IterativeImputer] Early stopping criterion not reached. warnings.warn( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\impute\_iterative.py:699: ConvergenceWarning: [IterativeImputer] Early stopping criterion not reached. warnings.warn( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\impute\_iterative.py:699: ConvergenceWarning: [IterativeImputer] Early stopping criterion not reached. warnings.warn( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\impute\_iterative.py:699: ConvergenceWarning: [IterativeImputer] Early stopping criterion not reached. warnings.warn( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\impute\_iterative.py:699: ConvergenceWarning: [IterativeImputer] Early stopping criterion not reached. warnings.warn( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\impute\_iterative.py:699: ConvergenceWarning: [IterativeImputer] Early stopping criterion not reached. warnings.warn(
| Prediction accuracy | Prediction Standard Deviation | |
|---|---|---|
| Clean Data | 0.866901 | 0.068551 |
| Simple fill w/ 0 | 0.851469 | 0.055465 |
| Simple Median | 0.856976 | 0.049874 |
| Simple Mean | 0.851505 | 0.048212 |
| Iterative w/ BayesianRidge | 0.855865 | 0.055839 |
| Iterative w/ Decision Tree | 0.852616 | 0.041811 |
| Iterative w/ ExtraTrees | 0.859126 | 0.049759 |
| Iterative with KN Reg | 0.853727 | 0.049938 |
| Cholesterol data dropped | 0.854778 | 0.055027 |
plot_imputers(x_labels,acc_mean,acc_std)
For simple and iterative imputers, the best score is most of the time 'Clean Data'. However, by fine tuning some of the classifiers, and adding other models to this list, it might be possible to add more accuracy to these predictions
train_imp=knn_df[knn_df.Cholesterol!=0].copy()
y_fulltrain_imp=train_imp.Cholesterol
X_fulltrain_imp=train_imp.drop('Cholesterol',axis=1)
miss_imp=knn_df[knn_df.Cholesterol==0].copy()
y_miss_imp=miss_imp.Cholesterol
X_miss_imp=miss_imp.drop('Cholesterol',axis=1)
The scikitlearn.impute class didn't yield satisfactory results, and all the advanced algorithms used were not fine tuned. Moreover, other algorithms can be used.
The model accuracy evaluation is done as followed:
import tensorflow as tf
import tensorflow.keras as keras
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Dropout
def manual_score(best_estimator,X_missing_chol=X_miss_imp,y_missing_chol=y_miss_imp,
X_clean=X_fulltrain_imp,y_clean=y_fulltrain_imp):
X_full=X_clean.append(X_missing_chol)
chol_pred=pd.Series(best_estimator.predict(scaler.fit_transform(X_missing_chol)),index=y_missing_chol.index)
chol_full_man=y_clean.append(chol_pred)
X_full['Cholesterol']=chol_full_man
y_full=X_full.HeartDisease
X_full.drop('HeartDisease',axis=1,inplace=True)
manual_impute_scores = cross_val_score(RandomForestClassifier(), scaler.fit_transform(X_full), y_full, scoring="accuracy", cv=20)
return manual_impute_scores
def imputer_random_tuning(imputer,param_grid,X_fulltrain_imp=X_fulltrain_imp,
y_fulltrain_imp=y_fulltrain_imp):
tuned_imputer = RandomizedSearchCV(estimator = imputer, param_distributions = param_grid,
n_iter = 100, cv = 5, n_jobs = -1)
tuned_imputer.fit(scaler.fit_transform(X_fulltrain_imp),y_fulltrain_imp)
return tuned_imputer
def nn_manual_score(best_estimator,X_missing_chol=X_miss_imp,y_missing_chol=y_miss_imp,
X_clean=X_fulltrain_imp,y_clean=y_fulltrain_imp):
X_full=X_clean.append(X_missing_chol)
chol_pred=pd.Series(best_estimator.predict(scaler.fit_transform(X_missing_chol)).reshape(y_miss_imp.index.shape),index=y_miss_imp.index)
chol_full_man=y_clean.append(chol_pred)
X_full['Cholesterol']=chol_full_man
y_full=X_full.HeartDisease
X_full.drop('HeartDisease',axis=1,inplace=True)
manual_impute_scores = cross_val_score(RandomForestClassifier(), scaler.fit_transform(X_full), y_full, scoring="accuracy", cv=20)
return manual_impute_scores
ext_reg=ExtraTreesRegressor()
random_grid = {
'n_estimators':np.linspace(2,100,20,dtype=int),
'max_depth': np.linspace(2,100,11,dtype=int),
'min_samples_split': np.linspace(2,50,18,dtype=int),
'min_samples_leaf': np.linspace(2,50,18,dtype=int),
'min_impurity_decrease':np.linspace(2,10,19,dtype=int),
'ccp_alpha':np.linspace(2,20,20),
'bootstrap': [True, False]
}
tuned_ext_reg=imputer_random_tuning(ext_reg,param_grid=random_grid)
print(tuned_ext_reg.best_params_)
{'n_estimators': 17, 'min_samples_split': 41, 'min_samples_leaf': 18, 'min_impurity_decrease': 4, 'max_depth': 90, 'ccp_alpha': 17.157894736842103, 'bootstrap': True}
ext_reg=ExtraTreesRegressor()
param_grid_ext = {'max_depth': [tuned_ext_reg.best_params_['max_depth']-10,
tuned_ext_reg.best_params_['max_depth'],
tuned_ext_reg.best_params_['max_depth']+10],
'min_samples_split': [tuned_ext_reg.best_params_['min_samples_split']-1,
tuned_ext_reg.best_params_['min_samples_split'],
tuned_ext_reg.best_params_['min_samples_split']+1],
'min_samples_leaf': [tuned_ext_reg.best_params_['min_samples_leaf']-1,
tuned_ext_reg.best_params_['min_samples_leaf'],
tuned_ext_reg.best_params_['min_samples_leaf']+1],
'min_impurity_decrease':[tuned_ext_reg.best_params_['min_impurity_decrease']-1,
tuned_ext_reg.best_params_['min_impurity_decrease'],
tuned_ext_reg.best_params_['min_impurity_decrease']+1],
'ccp_alpha':[tuned_ext_reg.best_params_['ccp_alpha']-0.1,
tuned_ext_reg.best_params_['ccp_alpha'],
tuned_ext_reg.best_params_['ccp_alpha']+0.1],
'bootstrap': [True]
}
ext_bp=GridSearchCV(ext_reg,param_grid=param_grid_ext,cv=10)
ext_bp.fit(scaler.fit_transform(X_fulltrain_imp),y_fulltrain_imp)
print(ext_bp.best_params_)
{'bootstrap': True, 'ccp_alpha': 17.257894736842104, 'max_depth': 80, 'min_impurity_decrease': 3, 'min_samples_leaf': 19, 'min_samples_split': 40}
ext_reg_score=manual_score(ext_bp.best_estimator_)
acc_mean=np.append(acc_mean,ext_reg_score.mean())
acc_std=np.append(acc_std,ext_reg_score.std())
x_labels=np.append(x_labels,'Tuned ExtraTreesRegressor')
print(ext_reg_score.mean())
0.8679468599033818
kn_reg=KNeighborsRegressor(n_jobs=-1)
random_grid = {'n_neighbors': [int(x) for x in np.linspace(1, 50, num = 50)],
'weights': ['uniform','distance'],
'algorithm': ['kd_tree','ball_tree'],
'p':np.linspace(1,10,10),
'leaf_size':[int(x) for x in np.linspace(2, 50, num = 15)]
}
tuned_kn_reg=imputer_random_tuning(imputer=kn_reg,param_grid=random_grid)
print(tuned_kn_reg.best_params_)
{'weights': 'distance', 'p': 5.0, 'n_neighbors': 49, 'leaf_size': 8, 'algorithm': 'kd_tree'}
knn_reg=KNeighborsRegressor(n_jobs=-1)
param_grid_knn = {'n_neighbors': [tuned_kn_reg.best_params_['n_neighbors']-1,
tuned_kn_reg.best_params_['n_neighbors'],
tuned_kn_reg.best_params_['n_neighbors']+1],
'weights': ['uniform','distance'],
'algorithm': ['kd_tree','ball_tree'],
'p':[tuned_kn_reg.best_params_['p']-1,
tuned_kn_reg.best_params_['p'],
tuned_kn_reg.best_params_['p']+1],
'leaf_size':[tuned_kn_reg.best_params_['leaf_size']-1,
tuned_kn_reg.best_params_['leaf_size'],
tuned_kn_reg.best_params_['leaf_size']+1]
}
knn_bp=GridSearchCV(knn_reg,param_grid=param_grid_knn,cv=10)
knn_bp.fit(scaler.fit_transform(X_fulltrain_imp),y_fulltrain_imp)
print(knn_bp.best_params_)
{'algorithm': 'kd_tree', 'leaf_size': 7, 'n_neighbors': 50, 'p': 6.0, 'weights': 'distance'}
kn_reg_score=manual_score(knn_bp.best_estimator_)
acc_mean=np.append(acc_mean,kn_reg_score.mean())
acc_std=np.append(acc_std,kn_reg_score.std())
x_labels=np.append(x_labels,'Tuned KNRegressor')
print(kn_reg_score.mean())
0.8635748792270531
xgb_reg=xgb.XGBRegressor()
random_grid = {'max_depth': [int(x) for x in np.linspace(1, 10, num = 5)],
'eta': [0.001,0.01,0.1,1],
'subsample':[float(x) for x in np.linspace(0.7, 1, num = 20)],
'colsample_bytree':[float(x) for x in np.linspace(0.7, 1, num = 20)]
}
tuned_xgb_reg=imputer_random_tuning(imputer=xgb_reg,param_grid=random_grid)
print(tuned_xgb_reg.best_params_)
{'subsample': 0.8894736842105263, 'max_depth': 1, 'eta': 0.1, 'colsample_bytree': 0.8578947368421053}
xgb_reg=xgb.XGBRegressor()
param_grid_xgb= {'max_depth': [tuned_xgb_reg.best_params_['max_depth']-1,
tuned_xgb_reg.best_params_['max_depth'],
tuned_xgb_reg.best_params_['max_depth']+1],
'eta': [tuned_xgb_reg.best_params_['eta']-0.05,
tuned_xgb_reg.best_params_['eta'],
tuned_xgb_reg.best_params_['eta']+0.05],
'subsample':[tuned_xgb_reg.best_params_['subsample']-0.01,
tuned_xgb_reg.best_params_['subsample'],
tuned_xgb_reg.best_params_['subsample']+0.01],
'colsample_bytree':[tuned_xgb_reg.best_params_['colsample_bytree']-0.05,
tuned_xgb_reg.best_params_['colsample_bytree'],
tuned_xgb_reg.best_params_['colsample_bytree']+0.05
]
}
xgb_bp=GridSearchCV(xgb_reg,param_grid=param_grid_xgb,cv=10)
xgb_bp.fit(scaler.fit_transform(X_fulltrain_imp),y_fulltrain_imp)
print(xgb_bp.best_params_)
{'colsample_bytree': 0.9078947368421053, 'eta': 0.05, 'max_depth': 2, 'subsample': 0.8794736842105263}
xgb_score=manual_score(xgb_bp.best_estimator_)
acc_mean=np.append(acc_mean,xgb_score.mean())
acc_std=np.append(acc_std,xgb_score.std())
x_labels=np.append(x_labels,'Tuned XGBoost Reg')
print(xgb_score.mean())
0.8625120772946862
Elastic Net being very quick to execute, there is no need to perform a RandomGridSearch on it
ela_reg=ElasticNet()
param_grid_xgb= {'l1_ratio': np.linspace(1e-5,1,100),
'alpha':np.linspace(1e-5,10,100)
}
ela_bp=GridSearchCV(ela_reg,param_grid=param_grid_xgb,cv=10)
ela_bp.fit(scaler.fit_transform(X_fulltrain_imp),y_fulltrain_imp)
print(ela_bp.best_params_)
C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.066e+06, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.719e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.157e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.089e+06, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.107e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.075e+06, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.148e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.174e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.155e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.131e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.064e+06, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.695e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.154e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.086e+06, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.104e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.072e+06, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.145e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.171e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.152e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.128e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.062e+06, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.671e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.151e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.083e+06, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.101e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.069e+06, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.142e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.168e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.149e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.125e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.060e+06, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.646e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.148e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.080e+06, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.098e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.066e+06, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.138e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.165e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.146e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.123e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.058e+06, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.620e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.145e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.077e+06, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.094e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.062e+06, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.135e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.162e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.143e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.120e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.055e+06, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.594e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.142e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.073e+06, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.091e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.059e+06, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.132e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.158e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.140e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.117e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.053e+06, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.568e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.139e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.070e+06, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.088e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.055e+06, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.128e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.155e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.137e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.114e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.051e+06, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.541e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.136e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.067e+06, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.085e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.052e+06, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.125e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.151e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.133e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.111e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.048e+06, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.513e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.133e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.063e+06, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.081e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.048e+06, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.121e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.148e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.130e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.108e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.046e+06, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.485e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.129e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.060e+06, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.078e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.044e+06, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.118e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.144e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.127e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.105e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.043e+06, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.455e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.126e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.056e+06, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.074e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.041e+06, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.114e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.140e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.123e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.101e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.040e+06, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.426e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.122e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.053e+06, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.070e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.037e+06, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.110e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.137e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.120e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.098e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.038e+06, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.395e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.119e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.049e+06, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.067e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.033e+06, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.106e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.133e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.116e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.095e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.035e+06, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.364e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.115e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.045e+06, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.063e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.029e+06, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.102e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.129e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.112e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.091e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.032e+06, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.332e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.111e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.041e+06, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.059e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.024e+06, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.098e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.125e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.109e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.088e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.029e+06, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.299e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.107e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.037e+06, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.055e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.020e+06, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.094e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.120e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.105e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.084e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.026e+06, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.266e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.103e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.033e+06, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.051e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.016e+06, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.090e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.116e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.101e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.080e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.023e+06, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.231e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.099e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.029e+06, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.046e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.011e+06, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.085e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.112e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.097e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.076e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.020e+06, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.196e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.095e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.024e+06, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.042e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.006e+06, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.081e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.107e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.092e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.072e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.017e+06, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.159e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.091e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.020e+06, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.037e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.002e+06, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.076e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.102e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.088e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.068e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.014e+06, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.122e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.087e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.015e+06, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.033e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.966e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.071e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.098e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.083e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.064e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.010e+06, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.084e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.082e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.010e+06, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.028e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.916e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.066e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.093e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.079e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.060e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.007e+06, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.044e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.077e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.006e+06, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.023e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.863e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.061e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.088e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.074e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.055e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.003e+06, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.004e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.073e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.001e+06, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.018e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.810e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.056e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.082e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.069e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.051e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.996e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.962e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.068e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.955e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.013e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.755e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.050e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.077e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.064e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.046e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.959e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.919e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.063e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.902e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.008e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.698e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.045e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.072e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.059e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.041e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.920e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.875e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.057e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.847e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.002e+06, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.640e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.039e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.066e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.054e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.036e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.880e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.830e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.052e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.791e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.966e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.580e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.033e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.060e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.049e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.031e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.839e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.783e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.047e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.734e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.908e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.518e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.027e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.054e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.043e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.026e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.797e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.735e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.041e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.674e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.849e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.454e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.021e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.048e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.037e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.021e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.754e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.685e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.035e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.613e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.787e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.388e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.015e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.042e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.031e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.015e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.709e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.634e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.029e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.549e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.723e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.320e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.008e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.035e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.025e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.009e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.663e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.581e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.023e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.484e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.658e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.250e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.001e+06, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.028e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.019e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.004e+06, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.615e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.526e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.016e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.416e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.590e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.177e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.943e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.021e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.012e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.975e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.565e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.470e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.010e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.347e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.520e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.103e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.870e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.014e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.006e+06, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.912e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.515e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.412e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.003e+06, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.275e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.448e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.025e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.795e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.007e+06, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.986e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.847e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.462e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.351e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.955e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.200e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.373e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.945e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.717e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.988e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.914e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.779e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.408e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.289e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.882e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.123e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.295e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.862e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.637e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.907e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.840e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.710e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.351e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.224e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.806e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.043e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.215e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.776e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.553e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.824e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.763e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.638e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.293e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.158e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.727e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.960e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.132e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.687e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.467e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.738e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.683e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.563e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.232e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.088e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.645e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.874e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.046e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.595e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.377e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.649e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.600e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.485e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.170e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.016e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.560e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.785e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.956e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.499e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.284e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.557e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.514e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.405e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.105e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.942e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.472e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.692e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.863e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.399e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.188e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.461e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.425e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.322e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.037e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.864e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.380e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.596e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.767e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.296e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.088e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.361e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.332e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.235e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.967e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.783e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.285e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.496e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.667e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.188e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.983e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.257e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.235e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.145e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.894e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.699e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.186e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.392e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.562e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.076e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.875e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.149e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.135e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.052e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.818e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.612e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.083e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.284e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.454e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.960e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.762e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.037e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.031e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.954e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.739e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.521e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.975e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.171e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.340e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.838e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.644e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.920e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.922e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.853e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.657e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.426e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.863e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.053e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.222e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.712e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.522e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.798e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.809e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.747e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.571e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.327e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.747e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.931e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.099e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.579e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.394e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.671e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.691e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.636e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.482e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.224e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.625e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.802e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.970e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.441e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.260e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.538e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.567e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.521e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.388e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.116e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.497e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.668e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.836e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.297e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.120e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.399e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.438e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.401e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.290e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.003e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.364e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.528e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.695e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.145e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.974e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.253e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.303e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.274e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.187e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.884e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.224e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.381e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.547e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.987e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.820e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.101e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.161e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.142e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.080e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.760e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.077e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.226e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.393e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.821e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.660e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.941e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.013e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.004e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.967e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.630e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.923e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.065e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.230e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.647e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.491e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.773e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.857e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.859e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.849e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.493e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.762e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.895e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.060e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.463e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.314e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.597e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.694e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.706e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.724e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.349e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.592e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.716e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.880e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.271e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.127e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.411e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.522e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.545e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.593e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.197e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.413e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.528e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.691e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.068e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.931e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.216e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.340e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.376e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.455e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.037e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.224e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.329e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.492e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.855e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.724e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.010e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.149e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.198e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.309e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.869e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.025e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.120e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.282e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.630e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.506e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.793e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.948e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.010e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.155e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.690e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.815e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.899e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.060e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.392e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.276e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.564e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.735e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.811e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.992e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.502e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.592e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.665e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.826e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.141e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.032e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.321e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.509e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.601e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.819e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.302e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.356e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.418e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.577e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.876e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.774e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.065e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.271e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.378e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.636e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.090e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.106e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.156e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.314e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.595e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.501e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.793e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.018e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.141e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.440e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.865e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.840e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.878e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.035e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.299e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.211e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.504e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.749e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.890e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.233e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.626e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.558e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.583e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.739e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.985e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.904e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.197e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.463e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.623e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.011e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.372e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.258e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.271e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.425e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.653e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.577e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.872e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.160e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.339e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.774e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.101e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.938e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.939e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.092e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.303e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.231e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.526e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.836e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.036e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.520e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.813e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.597e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.588e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.738e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.935e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.864e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.159e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.492e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.712e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.249e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.506e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.234e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.216e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.364e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.549e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.475e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.769e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.125e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.368e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.957e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.179e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.847e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.824e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.968e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.146e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.064e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.356e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.735e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.000e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.643e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.832e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.436e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.412e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.551e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.731e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.632e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.920e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.321e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.607e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.306e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.464e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.000e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.987e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.117e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.326e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.182e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.463e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.883e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.190e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.942e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.077e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.541e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.822e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.669e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.272e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.717e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.988e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.422e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.748e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.551e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.673e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.061e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.659e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.217e+05, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.219e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.250e+05, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.503e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.942e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.283e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.130e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.260e+05, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.568e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.499e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.794e+04, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.167e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.966e+04, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.023e+05, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.452e+05, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.799e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.680e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.499e+04, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.076e+05, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.340e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.861e+04, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.117e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.910e+04, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.748e+04, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.677e+04, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.307e+05, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.201e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.670e+04, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.114e+04, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.185e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.392e+03, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.068e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.204e+03, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.001e+04, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.189e+04, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.267e+04, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.698e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.560e+04, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.243e+04, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.035e+05, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.233e+03, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.020e+05, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.439e+03, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.429e+04, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.613e+04, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.942e+04, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.184e+05, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.224e+04, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.088e+04, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.887e+04, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.753e+03, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.737e+04, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.431e+03, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.912e+04, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.807e+03, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.575e+04, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.848e+04, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.498e+04, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.407e+04, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.486e+04, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.300e+03, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.286e+04, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.087e+03, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.451e+04, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.508e+03, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.168e+04, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.551e+04, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.784e+04, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.747e+04, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.156e+04, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.873e+03, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.848e+04, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.764e+03, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.049e+04, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.227e+03, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.182e+03, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.210e+03, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.079e+04, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.104e+04, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.908e+04, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.472e+03, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.422e+04, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.459e+03, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.081e+03, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.962e+03, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.297e+03, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.257e+04, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.382e+04, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.474e+04, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.756e+04, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.094e+03, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.009e+04, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.172e+03, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.308e+03, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.714e+03, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.821e+03, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.622e+04, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.689e+04, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.854e+04, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.718e+04, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.739e+03, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.608e+04, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.903e+03, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.044e+03, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.481e+03, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.374e+03, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.010e+04, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.001e+04, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.241e+04, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.109e+04, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.061e+03, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.218e+04, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.646e+03, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.788e+03, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.262e+03, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.956e+03, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.415e+04, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.314e+04, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.634e+04, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.529e+04, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.031e+04, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.841e+04, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.552e+03, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.547e+03, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.057e+03, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.566e+03, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.834e+04, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.629e+04, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.029e+04, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.964e+04, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.528e+04, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.476e+04, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.326e+04, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.322e+03, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.866e+03, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.201e+03, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.264e+04, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.943e+04, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.427e+04, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.408e+04, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.080e+04, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.122e+04, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.858e+04, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.402e+03, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.687e+03, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.060e+03, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.700e+04, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.258e+04, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.825e+04, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.859e+04, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.674e+04, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.810e+04, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.438e+04, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.393e+03, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.520e+03, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.026e+04, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.141e+04, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.570e+04, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.223e+04, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.315e+04, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.298e+04, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.099e+04, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.053e+04, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.492e+04, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.364e+03, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.667e+04, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.584e+04, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.881e+04, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.618e+04, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.773e+04, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.946e+04, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.385e+04, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.694e+04, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.258e+04, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.219e+03, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 2.397e+04, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.028e+04, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.189e+04, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.012e+04, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.231e+04, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.608e+04, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.667e+04, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.354e+04, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.107e+04, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 1.085e+03, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 3.191e+04, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.472e+04, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.495e+04, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.402e+04, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.689e+04, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.280e+04, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.945e+04, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.027e+04, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.013e+04, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 9.609e+02, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.031e+04, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.914e+04, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.797e+04, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.789e+04, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.145e+04, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.957e+04, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.219e+04, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.707e+04, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.960e+04, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.461e+02, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 4.903e+04, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.353e+04, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.096e+04, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.172e+04, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.598e+04, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.635e+04, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.489e+04, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.390e+04, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.932e+04, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.403e+02, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.794e+04, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.788e+04, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.391e+04, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.551e+04, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.047e+04, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.312e+04, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.754e+04, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.074e+04, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.918e+04, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.432e+02, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.696e+04, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.220e+04, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.683e+04, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.925e+04, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.491e+04, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.985e+04, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.016e+04, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.755e+04, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.909e+04, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 5.544e+02, tolerance: 2.429e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.601e+04, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.646e+04, tolerance: 2.301e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 6.970e+04, tolerance: 2.059e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.295e+04, tolerance: 2.437e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 7.930e+04, tolerance: 2.302e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.652e+04, tolerance: 2.346e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.274e+04, tolerance: 2.285e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.431e+04, tolerance: 2.407e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.900e+04, tolerance: 2.473e+02 model = cd_fast.enet_coordinate_descent( C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\linear_model\_coordinate_descent.py:647: ConvergenceWarning: Objective did not converge. You might want to increase the number of iterations, check the scale of the features or consider increasing regularisation. Duality gap: 8.505e+04, tolerance: 2.420e+02 model = cd_fast.enet_coordinate_descent(
{'alpha': 0.7070799999999999, 'l1_ratio': 1e-05}
elastic_score=manual_score(ela_bp.best_estimator_)
acc_mean=np.append(acc_mean,elastic_score.mean())
acc_std=np.append(acc_std,elastic_score.std())
x_labels=np.append(x_labels,'Tuned ElasticNet')
print(elastic_score.mean())
0.8592512077294685
Because of the time of execution, only one neural network will be trained. A further investigation could lead to an enhanced accuracy.
X_train_nn, X_test_nn, y_train_nn, y_test_nn=train_test_split(X_fulltrain_imp,y_fulltrain_imp)
X_train_nn=scaler.fit_transform(X_train_nn)
X_test_nn=scaler.fit_transform(X_test_nn)
#lr adaptation
lr_schedule = keras.optimizers.schedules.ExponentialDecay(
initial_learning_rate=1e-2,
decay_steps=10000,
decay_rate=0.9)
epochs=100
learning_rate = 0.0001
decay_rate = learning_rate / epochs
#Optimizer ADAM
adam = Adam(learning_rate=lr_schedule, decay=decay_rate)
input_dim = X_train_nn.shape[1]
#Neural Network 2 hidden layers of 16 neurons each, with dropout to avoid overfitting
lr_model = Sequential()
lr_model.add(Dense(16, activation=tf.nn.softmax, kernel_initializer='normal',
input_dim = input_dim))
lr_model.add(Dropout(0.2))
lr_model.add(Dense(16, activation=tf.nn.softmax, kernel_initializer='normal',
input_dim = input_dim))
lr_model.add(Dropout(0.5))
lr_model.add(Dense(1, kernel_initializer='normal', activation=tf.nn.relu))
# compile the model
lr_model.compile(loss=tf.keras.losses.MeanSquaredError(),optimizer=adam,metrics=['mean_absolute_percentage_error'])
#run the optimization
batch_size = int(input_dim/13)
lr_model_history = lr_model.fit(X_train_nn, y_train_nn,
batch_size=batch_size,
epochs=epochs,
verbose=1,
validation_data=(X_test_nn, y_test_nn))
Epoch 1/100 559/559 [==============================] - 2s 2ms/step - loss: 61237.8320 - mean_absolute_percentage_error: 97.7899 - val_loss: 57031.9727 - val_mean_absolute_percentage_error: 95.2426 Epoch 2/100 559/559 [==============================] - 1s 2ms/step - loss: 55700.2383 - mean_absolute_percentage_error: 92.4804 - val_loss: 52270.2500 - val_mean_absolute_percentage_error: 90.7369 Epoch 3/100 559/559 [==============================] - 1s 2ms/step - loss: 52139.0703 - mean_absolute_percentage_error: 89.0315 - val_loss: 48393.2305 - val_mean_absolute_percentage_error: 86.9046 Epoch 4/100 559/559 [==============================] - 1s 2ms/step - loss: 47792.7188 - mean_absolute_percentage_error: 84.6860 - val_loss: 44711.9023 - val_mean_absolute_percentage_error: 83.1115 Epoch 5/100 559/559 [==============================] - 1s 2ms/step - loss: 44998.4023 - mean_absolute_percentage_error: 81.8113 - val_loss: 41395.7773 - val_mean_absolute_percentage_error: 79.5487 Epoch 6/100 559/559 [==============================] - 1s 2ms/step - loss: 40973.7617 - mean_absolute_percentage_error: 77.3937 - val_loss: 38190.6992 - val_mean_absolute_percentage_error: 75.9560 Epoch 7/100 559/559 [==============================] - 1s 1ms/step - loss: 38717.6523 - mean_absolute_percentage_error: 74.4865 - val_loss: 35279.8867 - val_mean_absolute_percentage_error: 72.5487 Epoch 8/100 559/559 [==============================] - 1s 2ms/step - loss: 36306.3906 - mean_absolute_percentage_error: 71.5871 - val_loss: 32553.6387 - val_mean_absolute_percentage_error: 69.2153 Epoch 9/100 559/559 [==============================] - 1s 2ms/step - loss: 33322.9805 - mean_absolute_percentage_error: 67.1248 - val_loss: 29966.9414 - val_mean_absolute_percentage_error: 65.9073 Epoch 10/100 559/559 [==============================] - 1s 2ms/step - loss: 31155.5527 - mean_absolute_percentage_error: 64.2403 - val_loss: 27571.9297 - val_mean_absolute_percentage_error: 62.7002 Epoch 11/100 559/559 [==============================] - 1s 2ms/step - loss: 29475.6680 - mean_absolute_percentage_error: 61.3027 - val_loss: 25358.3711 - val_mean_absolute_percentage_error: 59.5942 Epoch 12/100 559/559 [==============================] - 1s 2ms/step - loss: 26935.7812 - mean_absolute_percentage_error: 57.1917 - val_loss: 23264.7090 - val_mean_absolute_percentage_error: 56.5117 Epoch 13/100 559/559 [==============================] - 1s 2ms/step - loss: 25559.9434 - mean_absolute_percentage_error: 55.6996 - val_loss: 21340.2520 - val_mean_absolute_percentage_error: 53.5348 Epoch 14/100 559/559 [==============================] - 1s 2ms/step - loss: 24904.6055 - mean_absolute_percentage_error: 53.3664 - val_loss: 19618.8555 - val_mean_absolute_percentage_error: 50.7514 Epoch 15/100 559/559 [==============================] - 1s 2ms/step - loss: 23455.8848 - mean_absolute_percentage_error: 50.7831 - val_loss: 17997.8887 - val_mean_absolute_percentage_error: 48.0423 Epoch 16/100 559/559 [==============================] - 1s 2ms/step - loss: 19994.1426 - mean_absolute_percentage_error: 46.5388 - val_loss: 16474.0508 - val_mean_absolute_percentage_error: 45.3967 Epoch 17/100 559/559 [==============================] - 1s 2ms/step - loss: 19858.1758 - mean_absolute_percentage_error: 46.8652 - val_loss: 15083.2900 - val_mean_absolute_percentage_error: 42.8858 Epoch 18/100 559/559 [==============================] - 1s 2ms/step - loss: 17813.6230 - mean_absolute_percentage_error: 42.6350 - val_loss: 13848.0771 - val_mean_absolute_percentage_error: 40.5602 Epoch 19/100 559/559 [==============================] - 1s 2ms/step - loss: 18391.5293 - mean_absolute_percentage_error: 42.7695 - val_loss: 12736.4736 - val_mean_absolute_percentage_error: 38.3516 Epoch 20/100 559/559 [==============================] - 1s 2ms/step - loss: 16679.4941 - mean_absolute_percentage_error: 41.1663 - val_loss: 11722.9785 - val_mean_absolute_percentage_error: 36.2258 Epoch 21/100 559/559 [==============================] - 1s 2ms/step - loss: 15534.0039 - mean_absolute_percentage_error: 39.1530 - val_loss: 10829.8447 - val_mean_absolute_percentage_error: 34.2480 Epoch 22/100 559/559 [==============================] - 1s 2ms/step - loss: 15902.7998 - mean_absolute_percentage_error: 40.7066 - val_loss: 10035.8672 - val_mean_absolute_percentage_error: 32.3933 Epoch 23/100 559/559 [==============================] - 1s 2ms/step - loss: 14366.5205 - mean_absolute_percentage_error: 37.5743 - val_loss: 9348.4355 - val_mean_absolute_percentage_error: 30.7611 Epoch 24/100 559/559 [==============================] - 1s 2ms/step - loss: 13980.1504 - mean_absolute_percentage_error: 36.5040 - val_loss: 8757.6641 - val_mean_absolute_percentage_error: 29.3335 Epoch 25/100 559/559 [==============================] - 1s 2ms/step - loss: 12477.0332 - mean_absolute_percentage_error: 35.5428 - val_loss: 8216.7598 - val_mean_absolute_percentage_error: 28.0303 Epoch 26/100 559/559 [==============================] - 1s 2ms/step - loss: 13410.8721 - mean_absolute_percentage_error: 38.1504 - val_loss: 7762.7969 - val_mean_absolute_percentage_error: 26.9242 Epoch 27/100 559/559 [==============================] - 1s 2ms/step - loss: 13460.2549 - mean_absolute_percentage_error: 36.0281 - val_loss: 7366.7983 - val_mean_absolute_percentage_error: 25.9520 Epoch 28/100 559/559 [==============================] - 1s 2ms/step - loss: 12774.4395 - mean_absolute_percentage_error: 35.7009 - val_loss: 7029.1299 - val_mean_absolute_percentage_error: 25.1345 Epoch 29/100 559/559 [==============================] - 1s 2ms/step - loss: 11558.0107 - mean_absolute_percentage_error: 34.4403 - val_loss: 6733.8999 - val_mean_absolute_percentage_error: 24.4051 Epoch 30/100 559/559 [==============================] - 1s 2ms/step - loss: 12384.3418 - mean_absolute_percentage_error: 35.6880 - val_loss: 6484.0459 - val_mean_absolute_percentage_error: 23.7848 Epoch 31/100 559/559 [==============================] - 1s 2ms/step - loss: 12108.5586 - mean_absolute_percentage_error: 35.3804 - val_loss: 6237.7739 - val_mean_absolute_percentage_error: 23.1582 Epoch 32/100 559/559 [==============================] - 1s 2ms/step - loss: 10961.2373 - mean_absolute_percentage_error: 34.2170 - val_loss: 6024.0542 - val_mean_absolute_percentage_error: 22.6148 Epoch 33/100 559/559 [==============================] - 1s 2ms/step - loss: 11097.4932 - mean_absolute_percentage_error: 33.3654 - val_loss: 6462.6157 - val_mean_absolute_percentage_error: 23.7124 Epoch 34/100 559/559 [==============================] - 1s 2ms/step - loss: 10437.9033 - mean_absolute_percentage_error: 31.8967 - val_loss: 6219.5293 - val_mean_absolute_percentage_error: 23.1471 Epoch 35/100 559/559 [==============================] - 1s 2ms/step - loss: 10188.1416 - mean_absolute_percentage_error: 31.1462 - val_loss: 6285.3027 - val_mean_absolute_percentage_error: 23.2834 Epoch 36/100 559/559 [==============================] - 1s 2ms/step - loss: 10096.8877 - mean_absolute_percentage_error: 31.0842 - val_loss: 7098.4839 - val_mean_absolute_percentage_error: 25.5337 Epoch 37/100 559/559 [==============================] - 1s 2ms/step - loss: 8704.3740 - mean_absolute_percentage_error: 28.2946 - val_loss: 5720.4795 - val_mean_absolute_percentage_error: 21.9611 Epoch 38/100 559/559 [==============================] - 1s 2ms/step - loss: 8760.7500 - mean_absolute_percentage_error: 28.2683 - val_loss: 5836.4004 - val_mean_absolute_percentage_error: 22.1981 Epoch 39/100 559/559 [==============================] - 1s 2ms/step - loss: 8502.3936 - mean_absolute_percentage_error: 28.0693 - val_loss: 5629.2476 - val_mean_absolute_percentage_error: 21.6092 Epoch 40/100 559/559 [==============================] - 1s 2ms/step - loss: 8010.7954 - mean_absolute_percentage_error: 26.1118 - val_loss: 5276.3052 - val_mean_absolute_percentage_error: 20.6663 Epoch 41/100 559/559 [==============================] - 1s 2ms/step - loss: 7475.1372 - mean_absolute_percentage_error: 25.8574 - val_loss: 5252.2656 - val_mean_absolute_percentage_error: 20.6088 Epoch 42/100 559/559 [==============================] - 1s 2ms/step - loss: 7086.2500 - mean_absolute_percentage_error: 24.4538 - val_loss: 5135.4907 - val_mean_absolute_percentage_error: 20.2589 Epoch 43/100 559/559 [==============================] - 1s 2ms/step - loss: 6751.4883 - mean_absolute_percentage_error: 24.1811 - val_loss: 5029.2681 - val_mean_absolute_percentage_error: 19.9225 Epoch 44/100 559/559 [==============================] - 1s 2ms/step - loss: 6669.0469 - mean_absolute_percentage_error: 23.8434 - val_loss: 4753.5527 - val_mean_absolute_percentage_error: 19.1545 Epoch 45/100 559/559 [==============================] - 1s 2ms/step - loss: 6481.4922 - mean_absolute_percentage_error: 22.9588 - val_loss: 4420.5908 - val_mean_absolute_percentage_error: 18.4106 Epoch 46/100 559/559 [==============================] - 1s 2ms/step - loss: 5983.9604 - mean_absolute_percentage_error: 22.4558 - val_loss: 4333.9028 - val_mean_absolute_percentage_error: 18.2041 Epoch 47/100 559/559 [==============================] - 1s 2ms/step - loss: 5527.3594 - mean_absolute_percentage_error: 21.4487 - val_loss: 3990.0081 - val_mean_absolute_percentage_error: 17.4217 Epoch 48/100 559/559 [==============================] - 1s 2ms/step - loss: 5447.3892 - mean_absolute_percentage_error: 21.5458 - val_loss: 3979.3264 - val_mean_absolute_percentage_error: 17.3821 Epoch 49/100 559/559 [==============================] - 1s 2ms/step - loss: 5196.6626 - mean_absolute_percentage_error: 20.9465 - val_loss: 3782.6162 - val_mean_absolute_percentage_error: 16.9499 Epoch 50/100 559/559 [==============================] - 1s 2ms/step - loss: 5170.0625 - mean_absolute_percentage_error: 20.8927 - val_loss: 3827.3804 - val_mean_absolute_percentage_error: 17.0413 Epoch 51/100 559/559 [==============================] - 1s 2ms/step - loss: 5117.8960 - mean_absolute_percentage_error: 20.7644 - val_loss: 3608.0432 - val_mean_absolute_percentage_error: 16.7078 Epoch 52/100 559/559 [==============================] - 1s 2ms/step - loss: 4748.1528 - mean_absolute_percentage_error: 20.5654 - val_loss: 3507.2524 - val_mean_absolute_percentage_error: 16.5407 Epoch 53/100 559/559 [==============================] - 1s 2ms/step - loss: 4721.6948 - mean_absolute_percentage_error: 20.3099 - val_loss: 3489.8186 - val_mean_absolute_percentage_error: 16.4963 Epoch 54/100 559/559 [==============================] - 1s 2ms/step - loss: 4809.4614 - mean_absolute_percentage_error: 20.1763 - val_loss: 3458.0471 - val_mean_absolute_percentage_error: 16.5064 Epoch 55/100 559/559 [==============================] - 1s 2ms/step - loss: 4714.7524 - mean_absolute_percentage_error: 19.7637 - val_loss: 3401.8647 - val_mean_absolute_percentage_error: 16.4528 Epoch 56/100 559/559 [==============================] - 1s 2ms/step - loss: 4310.5859 - mean_absolute_percentage_error: 19.6384 - val_loss: 3378.8867 - val_mean_absolute_percentage_error: 16.4812 Epoch 57/100 559/559 [==============================] - 1s 2ms/step - loss: 4627.4092 - mean_absolute_percentage_error: 19.5282 - val_loss: 3345.1323 - val_mean_absolute_percentage_error: 16.4480 Epoch 58/100 559/559 [==============================] - 1s 2ms/step - loss: 4370.5869 - mean_absolute_percentage_error: 19.1781 - val_loss: 3241.2354 - val_mean_absolute_percentage_error: 16.4127 Epoch 59/100 559/559 [==============================] - 1s 2ms/step - loss: 4320.8022 - mean_absolute_percentage_error: 19.4681 - val_loss: 3240.9893 - val_mean_absolute_percentage_error: 16.3929 Epoch 60/100 559/559 [==============================] - 1s 2ms/step - loss: 4234.1201 - mean_absolute_percentage_error: 19.2920 - val_loss: 3161.5342 - val_mean_absolute_percentage_error: 16.3252 Epoch 61/100 559/559 [==============================] - 1s 2ms/step - loss: 4489.7876 - mean_absolute_percentage_error: 20.1685 - val_loss: 3142.3545 - val_mean_absolute_percentage_error: 16.3778 Epoch 62/100 559/559 [==============================] - 1s 2ms/step - loss: 4325.9834 - mean_absolute_percentage_error: 19.7330 - val_loss: 3129.1326 - val_mean_absolute_percentage_error: 16.4153 Epoch 63/100 559/559 [==============================] - 1s 2ms/step - loss: 4179.1230 - mean_absolute_percentage_error: 19.5336 - val_loss: 3129.8816 - val_mean_absolute_percentage_error: 16.4367 Epoch 64/100 559/559 [==============================] - 1s 2ms/step - loss: 4046.0349 - mean_absolute_percentage_error: 18.8735 - val_loss: 3038.5000 - val_mean_absolute_percentage_error: 16.4724 Epoch 65/100 559/559 [==============================] - 1s 2ms/step - loss: 4222.2173 - mean_absolute_percentage_error: 19.6501 - val_loss: 3012.0364 - val_mean_absolute_percentage_error: 16.4007 Epoch 66/100 559/559 [==============================] - 1s 2ms/step - loss: 3942.6936 - mean_absolute_percentage_error: 19.0024 - val_loss: 3020.4202 - val_mean_absolute_percentage_error: 16.4624 Epoch 67/100 559/559 [==============================] - 1s 2ms/step - loss: 3895.2146 - mean_absolute_percentage_error: 19.1056 - val_loss: 3033.7427 - val_mean_absolute_percentage_error: 16.4451 Epoch 68/100 559/559 [==============================] - 1s 2ms/step - loss: 4084.9629 - mean_absolute_percentage_error: 19.4794 - val_loss: 2999.0962 - val_mean_absolute_percentage_error: 16.5397 Epoch 69/100 559/559 [==============================] - 1s 2ms/step - loss: 4152.4922 - mean_absolute_percentage_error: 19.8572 - val_loss: 2988.0034 - val_mean_absolute_percentage_error: 16.5649 Epoch 70/100 559/559 [==============================] - 1s 2ms/step - loss: 3771.3494 - mean_absolute_percentage_error: 18.7476 - val_loss: 2967.1633 - val_mean_absolute_percentage_error: 16.6220 Epoch 71/100 559/559 [==============================] - 1s 1ms/step - loss: 3992.6821 - mean_absolute_percentage_error: 19.2758 - val_loss: 2951.7319 - val_mean_absolute_percentage_error: 16.6115 Epoch 72/100 559/559 [==============================] - 1s 2ms/step - loss: 3993.6301 - mean_absolute_percentage_error: 19.5023 - val_loss: 2962.1848 - val_mean_absolute_percentage_error: 16.6127 Epoch 73/100 559/559 [==============================] - 1s 2ms/step - loss: 4018.5278 - mean_absolute_percentage_error: 19.7110 - val_loss: 2979.2871 - val_mean_absolute_percentage_error: 16.7320 Epoch 74/100 559/559 [==============================] - 1s 2ms/step - loss: 4056.6553 - mean_absolute_percentage_error: 19.7556 - val_loss: 2982.6838 - val_mean_absolute_percentage_error: 16.7053 Epoch 75/100 559/559 [==============================] - 1s 2ms/step - loss: 3911.3979 - mean_absolute_percentage_error: 19.1936 - val_loss: 2966.0291 - val_mean_absolute_percentage_error: 16.7689 Epoch 76/100 559/559 [==============================] - 1s 2ms/step - loss: 3828.1628 - mean_absolute_percentage_error: 19.1152 - val_loss: 2962.0034 - val_mean_absolute_percentage_error: 16.8432 Epoch 77/100 559/559 [==============================] - 1s 2ms/step - loss: 3781.0439 - mean_absolute_percentage_error: 18.9498 - val_loss: 2956.9812 - val_mean_absolute_percentage_error: 16.8213 Epoch 78/100 559/559 [==============================] - 1s 2ms/step - loss: 3961.4150 - mean_absolute_percentage_error: 19.7896 - val_loss: 2953.5327 - val_mean_absolute_percentage_error: 16.7860 Epoch 79/100 559/559 [==============================] - 1s 2ms/step - loss: 4072.5273 - mean_absolute_percentage_error: 19.5254 - val_loss: 2953.1650 - val_mean_absolute_percentage_error: 16.8345 Epoch 80/100 559/559 [==============================] - 1s 2ms/step - loss: 3972.3516 - mean_absolute_percentage_error: 19.5076 - val_loss: 2946.7100 - val_mean_absolute_percentage_error: 16.8462 Epoch 81/100 559/559 [==============================] - 1s 2ms/step - loss: 4027.8989 - mean_absolute_percentage_error: 19.6147 - val_loss: 2929.2239 - val_mean_absolute_percentage_error: 16.7912 Epoch 82/100 559/559 [==============================] - 1s 2ms/step - loss: 3889.4016 - mean_absolute_percentage_error: 19.5159 - val_loss: 2940.2185 - val_mean_absolute_percentage_error: 16.7852 Epoch 83/100 559/559 [==============================] - 1s 2ms/step - loss: 3970.8198 - mean_absolute_percentage_error: 19.6270 - val_loss: 2934.3552 - val_mean_absolute_percentage_error: 16.7637 Epoch 84/100 559/559 [==============================] - 1s 2ms/step - loss: 3851.9817 - mean_absolute_percentage_error: 19.2574 - val_loss: 2936.0615 - val_mean_absolute_percentage_error: 16.9206 Epoch 85/100 559/559 [==============================] - 1s 2ms/step - loss: 3903.7769 - mean_absolute_percentage_error: 19.7506 - val_loss: 2949.5256 - val_mean_absolute_percentage_error: 16.9332 Epoch 86/100 559/559 [==============================] - 1s 2ms/step - loss: 3720.3384 - mean_absolute_percentage_error: 19.1492 - val_loss: 2953.0171 - val_mean_absolute_percentage_error: 16.9778 Epoch 87/100 559/559 [==============================] - 1s 2ms/step - loss: 3694.1738 - mean_absolute_percentage_error: 19.2362 - val_loss: 2941.5317 - val_mean_absolute_percentage_error: 17.0412 Epoch 88/100 559/559 [==============================] - 1s 2ms/step - loss: 4009.1409 - mean_absolute_percentage_error: 19.8271 - val_loss: 2970.0500 - val_mean_absolute_percentage_error: 16.8724 Epoch 89/100 559/559 [==============================] - 1s 2ms/step - loss: 3923.4419 - mean_absolute_percentage_error: 19.4538 - val_loss: 2967.4341 - val_mean_absolute_percentage_error: 16.9123 Epoch 90/100 559/559 [==============================] - 1s 2ms/step - loss: 3854.9844 - mean_absolute_percentage_error: 19.4551 - val_loss: 2968.5781 - val_mean_absolute_percentage_error: 16.8875 Epoch 91/100 559/559 [==============================] - 1s 2ms/step - loss: 3796.3418 - mean_absolute_percentage_error: 19.0828 - val_loss: 2967.1975 - val_mean_absolute_percentage_error: 16.9172 Epoch 92/100 559/559 [==============================] - 1s 2ms/step - loss: 3789.5532 - mean_absolute_percentage_error: 19.2882 - val_loss: 2973.0750 - val_mean_absolute_percentage_error: 16.9546 Epoch 93/100 559/559 [==============================] - 1s 2ms/step - loss: 3874.0483 - mean_absolute_percentage_error: 19.2006 - val_loss: 2990.5193 - val_mean_absolute_percentage_error: 17.0198 Epoch 94/100 559/559 [==============================] - 1s 2ms/step - loss: 3823.6162 - mean_absolute_percentage_error: 19.3803 - val_loss: 2975.2810 - val_mean_absolute_percentage_error: 17.0281 Epoch 95/100 559/559 [==============================] - 1s 2ms/step - loss: 3832.3994 - mean_absolute_percentage_error: 19.5686 - val_loss: 2940.0396 - val_mean_absolute_percentage_error: 16.9079 Epoch 96/100 559/559 [==============================] - 1s 2ms/step - loss: 3671.8633 - mean_absolute_percentage_error: 19.0385 - val_loss: 2942.6013 - val_mean_absolute_percentage_error: 16.8567 Epoch 97/100 559/559 [==============================] - 1s 2ms/step - loss: 3829.8306 - mean_absolute_percentage_error: 19.4153 - val_loss: 2950.5374 - val_mean_absolute_percentage_error: 16.8363 Epoch 98/100 559/559 [==============================] - 1s 1ms/step - loss: 3748.9299 - mean_absolute_percentage_error: 19.3298 - val_loss: 2933.4585 - val_mean_absolute_percentage_error: 16.8520 Epoch 99/100 559/559 [==============================] - 1s 2ms/step - loss: 3888.3608 - mean_absolute_percentage_error: 19.5478 - val_loss: 2941.2854 - val_mean_absolute_percentage_error: 16.8518 Epoch 100/100 559/559 [==============================] - 1s 1ms/step - loss: 3863.6235 - mean_absolute_percentage_error: 19.3744 - val_loss: 2941.0183 - val_mean_absolute_percentage_error: 16.8098
# Plot the loss function
fig, ax = plt.subplots(1, 1, figsize=(10,6))
ax.plot(np.sqrt(lr_model_history.history['loss']), 'r', label='train')
ax.plot(np.sqrt(lr_model_history.history['val_loss']), 'b' ,label='val')
ax.set_xlabel(r'Epoch', fontsize=20)
ax.set_ylabel(r'MSE', fontsize=20)
ax.legend()
ax.set_title('MSE evolution over the epochs')
ax.tick_params(labelsize=20)
# Plot the MAE - metric
fig, ax = plt.subplots(1, 1, figsize=(10,6))
ax.plot(np.sqrt(lr_model_history.history['mean_absolute_percentage_error']), 'r', label='train')
ax.plot(np.sqrt(lr_model_history.history['val_mean_absolute_percentage_error']), 'b' ,label='val')
ax.set_xlabel(r'Epoch', fontsize=20)
ax.set_ylabel(r'MAE%', fontsize=20)
ax.set_title('MAE evolution over the epochs')
ax.legend()
ax.tick_params(labelsize=20)
As it is possible to see on the figures above, there are no big sign of overfitting. (The training data curve does not continue to decrease while the validation data curve stays stable)
nn_score=nn_manual_score(lr_model)
acc_mean=np.append(acc_mean,nn_score.mean())
acc_std=np.append(acc_std,nn_score.std())
x_labels=np.append(x_labels,'Simple NN')
print(nn_score.mean())
0.8668357487922705
plot_imputers(x_labels,acc_mean,acc_std)
The choice of imputer depends on the accuracy of the imputer. Due to the stochastic nature of the process, the imputer choice might differ from one run to another.
print('The best estimator is: '+str(x_labels[np.argmax(acc_mean)]))
The best estimator is: Tuned ExtraTreesRegressor
#get best estimator
best=np.argmax(acc_mean)
imp_df=original_df.copy()
X_full['HeartDisease']=y_full
X_full=X_fulltrain_imp.append(X_miss_imp)
base_strat=simple_strategy+iterative_strategy
custom_strat=[ext_bp.best_estimator_,knn_bp.best_estimator_,xgb_bp.best_estimator_,
ela_bp.best_estimator_,lr_model]
if best==0:
#imputer - clean data
imp_df=None
X=X_clean.copy()
y=y_clean.copy()
X['HeartDisease']=y
imp_df=original_df.loc[list(X_clean.index)].copy()
imp_df.Cholesterol=X.Cholesterol
elif 0<best<4:
#imputer - SimpleImputer
imputer = SimpleImputer(missing_values=np.nan,
strategy=base_strrat[best-1], fill_value=0, add_indicator=True)
new_chol=imputer.fit_transform(X_full)
imp_df['Cholesterol']=new_chol[:,2]
X=imp_df.iloc[:,:-1]
y=imp_df.iloc[:,-1]
elif 4<= best <=7:
#imputer - IterativeImputer
imputer = IterativeImputer(missing_values=np.nan, estimator=base_strat[best-1],
add_indicator=True)
new_chol=imputer.fit_transform(X_full)
imp_df['Cholesterol']=new_chol[:,2]
X=imp_df.iloc[:,:-1]
y=imp_df.iloc[:,-1]
elif 8<best<13:
#imputer- custom imputer
X_full=X_fulltrain_imp.append(X_miss_imp)
chol_pred=pd.Series(custom_strat[best-9].predict(scaler.fit_transform(X_miss_imp)),index=y_miss_imp.index)
chol_full_man=y_fulltrain_imp.append(chol_pred)
X_full['Cholesterol']=chol_full_man
X_full.sort_index(ascending=True,inplace=True)
imp_df['Cholesterol']=X_full['Cholesterol']
elif best==13:
#imputer - Neural Network
X_full=X_fulltrain_imp.append(X_miss_imp)
chol_pred=pd.Series(custom_strat[-1].predict(scaler.fit_transform(X_miss_imp)).reshape(y_miss_imp.index.shape),index=y_miss_imp.index)
chol_full_man=y_fulltrain_imp.append(chol_pred)
X_full['Cholesterol']=chol_full_man
X_full.sort_index(ascending=True,inplace=True)
imp_df['Cholesterol']=X_full['Cholesterol']
new_corr_mat=correlation_calculation(imp_df,data_type=data_type,plot_=True)
new_corr_mat
| Age | Sex | ChestPainType | RestingBP | Cholesterol | FastingBS | RestingECG | MaxHR | ExerciseAngina | Oldpeak | ST_Slope | HeartDisease | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Age | 1.000000 | 0.000000 | 0.083667 | 0.263084 | 0.052288 | 0.198170 | 0.078909 | -0.382280 | 0.216017 | 0.259482 | 0.084675 | 0.282012 |
| Sex | 0.000000 | 0.993448 | 0.040344 | 0.000000 | -0.107586 | 0.013750 | 0.000000 | -0.189668 | 0.035533 | 0.121682 | 0.023655 | 0.091463 |
| ChestPainType | 0.083667 | 0.040344 | 1.000000 | 0.000000 | 0.000000 | 0.025747 | 0.010548 | 0.176004 | 0.194978 | 0.099890 | 0.085683 | 0.293235 |
| RestingBP | 0.263084 | 0.000000 | 0.000000 | 1.000000 | 0.088173 | 0.067823 | 0.099792 | -0.109693 | 0.153064 | 0.147278 | 0.000000 | 0.117990 |
| Cholesterol | 0.052288 | -0.107586 | 0.000000 | 0.088173 | 1.000000 | 0.000000 | 0.000000 | -0.019212 | 0.081909 | 0.052070 | 0.463056 | 0.102180 |
| FastingBS | 0.198170 | 0.013750 | 0.025747 | 0.067823 | 0.000000 | 0.993914 | 0.016503 | -0.131067 | 0.000000 | 0.088961 | 0.031054 | 0.070438 |
| RestingECG | 0.078909 | 0.000000 | 0.010548 | 0.099792 | 0.000000 | 0.016503 | 1.000000 | 0.000000 | 0.011475 | 0.082396 | 0.000000 | 0.012071 |
| MaxHR | -0.382280 | -0.189668 | 0.176004 | -0.109693 | -0.019212 | -0.131067 | 0.000000 | 1.000000 | -0.370023 | -0.180418 | 0.202274 | -0.401410 |
| ExerciseAngina | 0.216017 | 0.035533 | 0.194978 | 0.153064 | 0.081909 | 0.000000 | 0.011475 | -0.370023 | 0.995478 | 0.409574 | 0.209528 | 0.243301 |
| Oldpeak | 0.259482 | 0.121682 | 0.099890 | 0.147278 | 0.052070 | 0.088961 | 0.082396 | -0.180418 | 0.409574 | 1.000000 | 0.249456 | 0.429377 |
| ST_Slope | 0.084675 | 0.023655 | 0.085683 | 0.000000 | 0.463056 | 0.031054 | 0.000000 | 0.202274 | 0.209528 | 0.249456 | 1.000000 | 0.387302 |
| HeartDisease | 0.282012 | 0.091463 | 0.293235 | 0.117990 | 0.102180 | 0.070438 | 0.012071 | -0.401410 | 0.243301 | 0.429377 | 0.387302 | 0.995593 |
The negative correlation between Cholesterol and HeartDisease became slightly positive.
The classifier used an untuned RandomForestRegressor to find the 'best imputer'. As a reminder, the accuracy was about (it varies because of the stochastic nature of the crossvalidation):
X=knn_df.loc[imp_df.index].copy()
y=X.HeartDisease
X.drop('HeartDisease',axis=1,inplace=True)
cross_val_score(RandomForestClassifier(),X,y,scoring='accuracy',error_score='raise',cv=20).mean()
0.8678019323671498
from sklearn.decomposition import PCA
X_scaled=scaler.fit_transform(X)
pca = PCA(n_components=20)
score=pca.fit_transform(X_scaled)
coeff=pca.components_ #eigen vectors.T
latent = pca.explained_variance_
explained = pca.explained_variance_ratio_
plt.plot(np.cumsum(explained))
plt.xticks(range(0,20,2))
plt.xlabel('number of components')
plt.ylabel('cumulative explained variance')
plt.title('Explained Variance with PCA decomposition')
plt.show()
The dataset used to perform PCA contains 20 features (11 initially, but 20 once encoded).
To reach a 100% explained variance, it is needed to get the 15 first components with linear PCA.
With only 9 components, 80% of the variance is explained.
Linear PCA linearity assumption makes it hard to perform decomposition if there are no existing linear combination between each of the feature. To explore non-linearity, PCA with kernel trick is used
from sklearn.decomposition import KernelPCA
strat=['cosine','rbf','poly']
fig,ax =plt.subplots(3,1,figsize=(12,18))
for i in range(0,len(strat)):
kpca = KernelPCA(kernel=strat[i],n_components=20)
k_score=kpca.fit_transform(X_scaled) #eigen vectors.T
explained_variance = np.var(k_score, axis=0)
explained_variance_ratio = explained_variance / np.sum(explained_variance)
ax[i].plot(np.cumsum(explained_variance_ratio),label='Kernel : '+strat[i])
ax[i].set_xticks(range(0,20,2))
ax[i].set_xlabel('number of components')
ax[i].set_ylabel('cumulative explained variance')
ax[i].legend()
fig.suptitle('Explained Variance for different Kernel in KernelPCA decomposition',y=0.9)
plt.show()
On the graph above it seems that only Cosine Kernel manages to explain 100% Kernel in less than 20 components
Regarding 80% explained variance, cosine is very similar to linear PCA, reaching it in 9 components. However rbf and polynomial kernel reaches it in 12 components
It is possible to find the best KernelPCA kernel and hyperparameters by doing a RandomGridSearch.
Optimization for Kernel PCA:
#gain function is the cumsum explained variance to the 4th component
def score_pca(estimator,X):
X_score=estimator.transform(X)
explained_variance = np.var(X_score, axis=0)
explained_variance_ratio = explained_variance / np.sum(explained_variance)
score=np.cumsum(explained_variance_ratio)
return score[3]
kpca=KernelPCA()
random_grid={'gamma':np.linspace(0,1,50),
'degree':[2,3,4],
'coef0':np.linspace(0,10000,1000),
'kernel':['rbf','poly','cosine']}
cv=RandomizedSearchCV(kpca,param_distributions=random_grid,n_iter=5000,n_jobs=-1,scoring=score_pca)
cv.fit(X_scaled)
print('Best score for a non-linear Kernel PCA: '+str(cv.best_score_))
print('Best estimation of a linear PCA: '+str(np.cumsum(pca.explained_variance_ratio_)[3]))
C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\model_selection\_validation.py:372: FitFailedWarning:
43 fits failed out of a total of 25000.
The score on these train-test partitions for these parameters will be set to nan.
If these failures are not expected, you can try to debug them by setting error_score='raise'.
Below are more details about the failures:
--------------------------------------------------------------------------------
14 fits failed with the following error:
Traceback (most recent call last):
File "C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\model_selection\_validation.py", line 678, in _fit_and_score
estimator.fit(X_train, **fit_params)
File "C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\decomposition\_kernel_pca.py", line 432, in fit
self._fit_transform(K)
File "C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\decomposition\_kernel_pca.py", line 360, in _fit_transform
self.eigenvalues_ = _check_psd_eigenvalues(
File "C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\utils\validation.py", line 1475, in _check_psd_eigenvalues
raise ValueError(
ValueError: There are significant negative eigenvalues (1.82513e+14 of the maximum positive). Either the matrix is not PSD, or there was an issue while computing the eigendecomposition of the matrix.
--------------------------------------------------------------------------------
27 fits failed with the following error:
Traceback (most recent call last):
File "C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\model_selection\_validation.py", line 678, in _fit_and_score
estimator.fit(X_train, **fit_params)
File "C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\decomposition\_kernel_pca.py", line 432, in fit
self._fit_transform(K)
File "C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\decomposition\_kernel_pca.py", line 360, in _fit_transform
self.eigenvalues_ = _check_psd_eigenvalues(
File "C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\utils\validation.py", line 1475, in _check_psd_eigenvalues
raise ValueError(
ValueError: There are significant negative eigenvalues (5.78631e+14 of the maximum positive). Either the matrix is not PSD, or there was an issue while computing the eigendecomposition of the matrix.
--------------------------------------------------------------------------------
2 fits failed with the following error:
Traceback (most recent call last):
File "C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\model_selection\_validation.py", line 678, in _fit_and_score
estimator.fit(X_train, **fit_params)
File "C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\decomposition\_kernel_pca.py", line 432, in fit
self._fit_transform(K)
File "C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\decomposition\_kernel_pca.py", line 360, in _fit_transform
self.eigenvalues_ = _check_psd_eigenvalues(
File "C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\utils\validation.py", line 1475, in _check_psd_eigenvalues
raise ValueError(
ValueError: There are significant negative eigenvalues (5.3784e+14 of the maximum positive). Either the matrix is not PSD, or there was an issue while computing the eigendecomposition of the matrix.
warnings.warn(some_fits_failed_message, FitFailedWarning)
C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\model_selection\_search.py:969: UserWarning: One or more of the test scores are non-finite: [0.44093142 0.10009618 0.44093142 ... 0.09521862 0.44093142 0.40359815]
warnings.warn(
Best score for a non-linear Kernel PCA: 0.4409314217816404 Best estimation of a linear PCA: 0.4878953372423539
pca=PCA(n_components=10)
pc=pca.fit_transform(X_scaled)
pc1=pc[:,0]
pc2=pc[:,1]
pc3=pc[:,2]
kpca=KernelPCA(n_components=10,kernel='cosine')
kpc=kpca.fit_transform(X_scaled)
kpc1=kpc[:,0]
kpc2=kpc[:,1]
kpc3=kpc[:,2]
fig,ax=plt.subplots(2,1,figsize=(12,16))
ax.flat
colour = ['r' if y == 1 else 'b' for y in y]
ax[0].scatter(pc1,pc2,c=colour,edgecolors='#000000')
ax[0].set_ylabel("PC2",size=20)
ax[0].set_ylabel("PC1",size=20)
ax[0].set_title('Linear PCA decomposition first two components')
ax[1].scatter(kpc1,kpc2,c=colour,edgecolors='#000000')
ax[1].set_ylabel("PC2",size=20)
ax[1].set_ylabel("PC1",size=20)
ax[1].set_title('Kernel PCA decomposition first two components using cosine Kernel')
plt.show()
This is a representation of the PCA decomposition in 2D for Cosine Kernel PCA and Linear PCA.
The explained variance with 4 components is 49.83% for the linear PCA, and 47.00% for the Kernel PCA with cosine function.
Another option for dimensionality reduction would be to use Multiple Correspondence Analysis, which is PCA applied to categorical variables.
Another technique called Factor Analysis of Mixed Data (FAMD) combines Principal Component Analysis (PCA) to deal with numerical variables and Multiple Correspondence Analysis (MCA) for categorical ones.
To use MCA, it is needed to have a fully categorical data set. The strategy to convert the numerical data to categorical is:
X_cat=imp_df.iloc[:,:-1].copy()
age_cat=[]
for age in X_cat['Age']:
if age<=40:
age_cat.append('Low')
elif 60<=age>40:
age_cat.append('Medium')
else:
age_cat.append('High')
X_cat['Age']=age_cat
bp_cat=[]
for bp in X_cat['RestingBP']:
if bp<90:
bp_cat.append('Low')
elif 90<=age>140:
bp_cat.append('Normal')
else:
bp_cat.append('High')
X_cat['RestingBP']=bp_cat
chol_cat=[]
for chol in X_cat['Cholesterol']:
if chol<120:
chol_cat.append('Low')
elif 120<=chol>220:
chol_cat.append('Normal')
else:
chol_cat.append('High')
X_cat['Cholesterol']=chol_cat
diab_cat=[]
for diab in X_cat['FastingBS']:
if diab==1:
diab_cat.append('Yes')
else:
diab_cat.append('No')
X_cat['FastingBS']=diab_cat
hr_cat=[]
estimated_hr=220-imp_df['Age']-imp_df['MaxHR']
for hr in estimated_hr:
if hr<0:
hr_cat.append('Alarming')
else:
hr_cat.append('Normal')
X_cat['MaxHR']=hr_cat
op_cat=[]
for op in X_cat['Oldpeak'].astype(int):
if op==0:
op_cat.append('No')
else:
op_cat.append('Yes')
X_cat['Oldpeak']=op_cat
import prince
mca=prince.MCA(n_components=4)
mca.fit(X_cat)
print('MCA - Explained variance with 4 components: '+str(np.cumsum(mca.explained_inertia_)[-1]))
MCA - Explained variance with 4 components: 0.37111507429630414
Transforming the dataset into a categorical dataset does not yield a better explained variance. MCA with 4 components reaches 40.83% explained variance. It can be explained by the loss of information due to transformation from num-cat and the 'arbitrary' values to split the different categories.
X_ori=imp_df.iloc[:,:-1].copy()
X_ori.iloc[:,[0,3,4,7,9]]=scaler.fit_transform(X_ori.iloc[:,[0,3,4,7,9]])
import prince
famd = prince.FAMD(n_components=4,n_iter=100,copy=True,check_input=True,engine='auto')
famd.fit(X_ori)
FAMD(n_components=4, n_iter=100)
plt.bar(['PC1','PC2','PC3','PC4','Total'],np.append(famd.explained_inertia_,np.cumsum(famd.explained_inertia_)[-1]))
plt.ylabel('Explained Variance')
plt.xlabel('Principal Component')
plt.title('Explained Variance of the 4 first principal components')
plt.show()
print('FAMD - Explained variance with 4 components: '+str(np.cumsum(famd.explained_inertia_)[-1]))
FAMD - Explained variance with 4 components: 0.4715994313324581
famd.column_correlations(X_ori)
| 0 | 1 | 2 | 3 | |
|---|---|---|---|---|
| Age | 0.744713 | 0.473113 | -0.434988 | 0.087509 |
| ChestPainType_ASY | 0.381785 | 0.029912 | 0.010625 | -0.038128 |
| ChestPainType_ATA | -0.347889 | -0.046267 | 0.066804 | -0.058270 |
| ChestPainType_NAP | -0.126176 | -0.017714 | -0.055254 | 0.042519 |
| ChestPainType_TA | -0.008494 | 0.048270 | -0.039085 | 0.110755 |
| Cholesterol | 0.168305 | 0.607847 | 0.753955 | -0.611174 |
| ExerciseAngina_N | -0.475324 | -0.096879 | -0.025298 | -0.090609 |
| ExerciseAngina_Y | 0.475324 | 0.096879 | 0.025298 | 0.090609 |
| FastingBS | 0.282824 | 0.178592 | -0.314182 | -0.238845 |
| MaxHR | -0.692026 | 0.032978 | 0.301067 | 0.287299 |
| Oldpeak | 0.646493 | 0.110746 | 0.203397 | 0.550412 |
| RestingBP | 0.444747 | 0.752712 | -0.155654 | 0.271627 |
| RestingECG_LVH | 0.055259 | 0.162251 | 0.046968 | 0.092817 |
| RestingECG_Normal | -0.193324 | -0.179921 | 0.071909 | -0.080260 |
| RestingECG_ST | 0.182965 | 0.057160 | -0.136978 | 0.004635 |
| ST_Slope_Down | 0.238146 | 0.004629 | 0.011532 | 0.180125 |
| ST_Slope_Flat | 0.396692 | 0.092191 | 0.011100 | -0.016017 |
| ST_Slope_Up | -0.522200 | -0.095453 | -0.017099 | -0.075838 |
| Sex_F | -0.164595 | 0.090227 | 0.137425 | -0.042717 |
| Sex_M | 0.164595 | -0.090227 | -0.137425 | 0.042717 |
ax = famd.plot_row_coordinates(X_ori,ax=None,figsize=(12, 16),x_component=0,y_component=1,
color_labels=['HeartDisease {}'.format(t) for t in y],
ellipse_outline=False,ellipse_fill=True,show_points=True)
As we can see, decomposition using PCA and MCA combined results in a similar explained variance than the linear PCA.
Now that the imputing method has been selected, data can be used to build models. A non-exhaustive list of models has been elaborated:
from sklearn.metrics import confusion_matrix
from sklearn.metrics import accuracy_score
X_pca=PCA(n_components=20).fit_transform(X_scaled)
X_kpca=KernelPCA(kernel='cosine',n_components=20).fit_transform(X_scaled)
X_famd=prince.FAMD(n_components=20).fit_transform(X_ori)
X_train_ori,y_train,X_test_ori,y_test,X_val_ori,y_val=get_train_test_val(X,y,0.3)
X_test=scaler.fit_transform(X_test_ori)
X_train=scaler.fit_transform(X_train_ori)
X_val=scaler.fit_transform(X_val_ori)
X_train_pca,y_train_pca,X_test_pca,y_test_pca,X_val_pca,y_val_pca=get_train_test_val(X_pca,y,0.3)
X_train_kpca,y_train_kpca,X_test_kpca,y_test_kpca,X_val_kpca,y_val_kpca=get_train_test_val(X_kpca,y,0.3)
X_train_famd,y_train_famd,X_test_famd,y_test_famd,X_val_famd,y_val_famd=get_train_test_val(X_famd,y,0.3)
X_train_famd=np.array(X_train_famd)
y_train_famd=np.array(y_train_famd)
X_test_famd=np.asarray(X_test_famd)
y_test_famd=np.asarray(y_test_famd)
X_val_famd=np.asarray(X_val_famd)
y_val_famd=np.asarray(y_val_famd)
from sklearn.preprocessing import MinMaxScaler,RobustScaler,PowerTransformer,normalize
# automatic svm hyperparameter tuning using skopt for the ionosphere dataset
from pandas import read_csv
from sklearn.model_selection import cross_val_score
from sklearn.svm import SVC
from sklearn.model_selection import RepeatedStratifiedKFold
from skopt import BayesSearchCV
#get Accuracy and ROC_AUC for the validatation and the test set, aswell as the confusion matrix
def clf_score(estimator,X_val,y_val,X_test,y_test,cv,summary=True):
cv_=cross_val_score(estimator,X_val,y_val,scoring='roc_auc',cv=cv)
if summary==True:
print('Roc_AUC validation set: '+str(cv_.mean()))
print('ROC_AUC test set: '+str(roc_auc_score(estimator.predict(X_test),y_test)))
print('Accuracy validation set:'+str(accuracy_score(estimator.predict(X_val),y_val)))
print('Accuracy test set:'+str(accuracy_score(estimator.predict(X_test),y_test)))
print('Confusion Matrix:')
print(confusion_matrix(y_test,estimator.predict(X_test)))
else:
pass
return cv_,roc_auc_score(estimator.predict(X_test),y_test),accuracy_score(estimator.predict(X_val),y_val),accuracy_score(estimator.predict(X_test),y_test)
#display ROC and Accuracy for different feature scaling methods
def transformation_comp(name,estimator,X_train=X_train_ori,X_test=X_test_ori,X_val=X_val_ori,
y_train=y_train,y_test=y_test,y_val=y_val):
stdscaler=StandardScaler()
minmaxscaler=MinMaxScaler()
robustscaler=RobustScaler()
power=PowerTransformer()
scalers=[stdscaler,minmaxscaler,robustscaler,power]
scalers_name=['Standard Scaling','MinMax Scaling','Robust Scaling','Power Transformation']
val_acc=[]
test_acc=[]
val_roc=[]
test_roc=[]
for scaler in scalers:
print(scaler)
X_t=scaler.fit_transform(X_train)
X_tt=scaler.fit_transform(X_test)
X_v=scaler.fit_transform(X_val)
estimator.fit(X_t,y_train)
scores=clf_score(estimator,X_v,y_val,X_tt,y_test,cv=20)
val_roc.append(scores[0].mean())
test_roc.append(scores[1])
val_acc.append(scores[2])
test_acc.append(scores[3])
length = len(test_acc)
width = 0.2
x = np.arange(length)
fig,ax=plt.subplots(1)
ax.bar(x, val_roc, width, color='b', label='Validation ROC')
ax.bar(x + width, test_roc, width, color='r', label='Test ROC')
ax.bar(x + (2 * width), val_acc, width, color='y', label='Validation Accuracy')
ax.bar(x + (3 * width), test_acc, width, color='g', label='Test Accuracy')
ax.set_xticks(x + width + width/2)
ax.set_xticklabels(scalers_name)
ax.set_ylabel('Score')
ax.set_ylim(0.7,1)
ax.set_xlabel('Feature Transformation Strategy')
ax.set_title('Comparison of prediction accuracy using '+name+' for Heart Disease Prediction')
ax.legend()
plt.grid(True, 'major', 'y', ls='--', lw=.5, c='k', alpha=.3)
#Compare an estimator using PCA/KernelPCA using cosine/FAMD and original data
def plot_perf_comp(name,best_estimator_,ori_val_roc,ori_test_roc,ori_val_acc,ori_test_acc,
X_train_pca=X_train_pca,y_train_pca=y_train_pca, X_test_pca=X_test_pca,
y_test_pca=y_test_pca,X_val_pca=X_val_pca, y_val_pca=y_val_pca,
X_train_famd=X_train_famd,y_train_famd=y_train_famd, X_test_famd=X_test_famd,
y_test_famd=y_test_famd,X_val_famd=X_val_famd,y_val_famd=y_val_famd,
X_train_kpca=X_train_kpca, y_train_kpca=y_train_kpca,X_val_kpca=X_val_kpca,
y_val_kpca=y_val_kpca, X_test_kpca=X_test_kpca,y_test_kpca=y_test_kpca):
n_components=np.linspace(1,20,20).astype(int)
famd_test_score=[]
famd_val_score=[]
pca_test_score=[]
pca_val_score=[]
kpca_test_score=[]
kpca_val_score=[]
pca_test_acc=[]
pca_val_acc=[]
famd_test_acc=[]
famd_val_acc=[]
kpca_test_acc=[]
kpca_val_acc=[]
for n in n_components:
clf1=best_estimator_
clf2=best_estimator_
clf3=best_estimator_
clf1.fit(X_train_pca[:,:n],y_train_pca)
score1=clf_score(clf1,X_val_pca[:,:n],y_val_pca,X_test_pca[:,:n],y_test_pca,cv=5,summary=False)
clf2.fit(X_train_famd[:,:n],y_train_famd)
score2=clf_score(clf2,X_val_famd[:,:n],y_val_famd,X_test_famd[:,:n],y_test_famd,cv=5,summary=False)
clf3.fit(X_train_kpca[:,:n],y_train_kpca)
score3=clf_score(clf3,X_val_kpca[:,:n],y_val_kpca,X_test_kpca[:,:n],y_test_kpca,cv=5,summary=False)
pca_test_score.append(score1[1])
pca_val_score.append(score1[0].mean())
pca_val_acc.append(score1[2])
pca_test_acc.append(score1[3])
famd_test_score.append(score2[1])
famd_val_score.append(score2[0].mean())
famd_val_acc.append(score2[2])
famd_test_acc.append(score2[3])
kpca_test_score.append(score3[1])
kpca_val_score.append(score3[0].mean())
kpca_val_acc.append(score3[2])
kpca_test_acc.append(score3[3])
fig,ax=plt.subplots(2,1,figsize=(12,18))
ax.flat
ax[0].plot(n_components,pca_test_score,label='PCA_test (AUC='+str(round(max(pca_test_score),3))+')')
ax[0].plot(n_components,pca_val_score,label='PCA_val (AUC='+str(round(max(pca_val_score),3))+')')
ax[0].plot(n_components,famd_val_score,label='FAMD_val (AUC='+str(round(max(famd_val_score),3))+')')
ax[0].plot(n_components,famd_test_score,label='FAMD_test (AUC='+str(round(max(famd_test_score),3))+')')
ax[0].plot(n_components,np.full(len(n_components),ori_val_roc),label='Original_val (AUC='+str(round(ori_val_roc,3))+')',
linestyle='dashed')
ax[0].plot(n_components,np.full(len(n_components),ori_test_roc),label='Original_test (AUC='+str(round(ori_test_roc,3))+')',
linestyle='dashed')
ax[0].plot(n_components,kpca_test_score,label='KernelPCA_test (AUC='+str(round(max(kpca_test_score),3))+')')
ax[0].plot(n_components,kpca_val_score,label='KernelPCA_val (AUC='+str(round(max(kpca_val_score),3))+')')
ax[0].legend(loc='lower right')
ax[0].set_xlabel('Number of components')
ax[0].set_ylabel('ROC_AUC Score')
ax[0].set_title('Comparison of ROC_AUC score using '+name+' for Heart Disease Prediction')
ax[1].plot(n_components,pca_test_acc,label='PCA_test (acc='+str(round(max(pca_test_acc),3))+')')
ax[1].plot(n_components,pca_val_acc,label='PCA_val (acc='+str(round(max(pca_val_acc),3))+')')
ax[1].plot(n_components,famd_val_acc,label='FAMD_val (acc='+str(round(max(famd_val_acc),3))+')')
ax[1].plot(n_components,famd_test_acc,label='FAMD_test (acc='+str(round(max(famd_test_acc),3))+')')
ax[1].plot(n_components,np.full(len(n_components),ori_val_acc),label='Original_val (acc='+str(round(ori_val_acc,3))+')',
linestyle='dashed')
ax[1].plot(n_components,np.full(len(n_components),ori_test_acc),label='Original_test (acc='+str(round(ori_test_acc,3))+')',
linestyle='dashed')
ax[1].plot(n_components,kpca_test_acc,label='KernelPCA_test (acc='+str(round(max(kpca_test_acc),3))+')')
ax[1].plot(n_components,kpca_val_acc,label='KernelPCA_val (acc='+str(round(max(kpca_val_acc),3))+')')
ax[1].legend(loc='lower right')
ax[1].set_xlabel('Number of components')
ax[1].set_ylabel('Accuracy')
ax[1].set_title('Comparison of prediction accuracy using '+name+' for Heart Disease Prediction')
return famd_test_acc,pca_test_acc,kpca_test_acc
For each classifier a comparison of the scaling impact is performed, as well as an accuracy comparison using different data (PCA,FAMD,KernelPCA and original). Two metrics are used to compare models’ performance – Accuracy and ROC-AUC score
from sklearn.linear_model import LogisticRegression
log_clf=LogisticRegression(max_iter=1000,n_jobs=-1)
param_grid_lr={
'solver':['newton-cg', 'lbfgs'],
'penalty':['l2'],
'C':np.linspace(1e-3,1e-1,100)
}
log_bp=GridSearchCV(log_clf,param_grid=param_grid_lr,cv=20)
log_bp.fit(X_train,y_train)
print(log_bp.best_params_)
log_cv,log_score,log_acc_valscore,log_acc_testscore=clf_score(log_bp.best_estimator_,X_val,y_val,X_test,y_test,cv=20)
{'C': 0.018000000000000002, 'penalty': 'l2', 'solver': 'newton-cg'}
Roc_AUC validation set: 0.9141666666666666
ROC_AUC test set: 0.8644334654432922
Accuracy validation set:0.8497409326424871
Accuracy test set:0.8623188405797102
Confusion Matrix:
[[ 99 24]
[ 14 139]]
log_famd_acc,log_pca_acc,log_kpca_acc=plot_perf_comp('Logistic Regression',log_bp.best_estimator_,log_cv.mean(),log_score,log_acc_valscore,log_acc_testscore)
transformation_comp('Logistic Regression',log_bp.best_estimator_)
StandardScaler() Roc_AUC validation set: 0.9141666666666666 ROC_AUC test set: 0.8644334654432922 Accuracy validation set:0.8497409326424871 Accuracy test set:0.8623188405797102 Confusion Matrix: [[ 99 24] [ 14 139]] MinMaxScaler() Roc_AUC validation set: 0.9010833333333332 ROC_AUC test set: 0.8810765815760268 Accuracy validation set:0.844559585492228 Accuracy test set:0.8731884057971014 Confusion Matrix: [[ 97 26] [ 9 144]] RobustScaler() Roc_AUC validation set: 0.9195 ROC_AUC test set: 0.8378151260504201 Accuracy validation set:0.8341968911917098 Accuracy test set:0.8369565217391305 Confusion Matrix: [[109 14] [ 31 122]] PowerTransformer()
C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\preprocessing\_data.py:3253: RuntimeWarning: divide by zero encountered in log loglike = -n_samples / 2 * np.log(x_trans.var())
Roc_AUC validation set: 0.9097500000000001 ROC_AUC test set: 0.8559006211180126 Accuracy validation set:0.8497409326424871 Accuracy test set:0.855072463768116 Confusion Matrix: [[ 99 24] [ 16 137]]
from sklearn.svm import SVC
svm_clf=SVC(max_iter=-1,probability=True)
param_grid_svm={
'gamma':[0.001,0.01,0.05,0.1,0.5],
'kernel':['linear','rbf','sigmoid'],
'C':[0.01,0.05,0.1,0.5]
}
svm_bp=GridSearchCV(svm_clf,param_grid=param_grid_svm,cv=20)
svm_bp.fit(X_train,y_train)
print(svm_bp.best_params_)
svm_cv,svm_score,svm_acc_v,svm_acc_t=clf_score(svm_bp.best_estimator_,X_val,y_val,X_test,y_test,cv=20)
{'C': 0.5, 'gamma': 0.001, 'kernel': 'linear'}
Roc_AUC validation set: 0.9078333333333333
ROC_AUC test set: 0.8583458485303583
Accuracy validation set:0.8186528497409327
Accuracy test set:0.8586956521739131
Confusion Matrix:
[[101 22]
[ 17 136]]
svm_famd_acc,svm_pca_acc,svm_kpca_acc=plot_perf_comp('Support Vector Machine',svm_bp.best_estimator_,svm_cv.mean(),svm_score,svm_acc_v,svm_acc_t)
transformation_comp('Support Vector Machine',svm_bp.best_estimator_)
StandardScaler() Roc_AUC validation set: 0.9078333333333333 ROC_AUC test set: 0.8583458485303583 Accuracy validation set:0.8186528497409327 Accuracy test set:0.8586956521739131 Confusion Matrix: [[101 22] [ 17 136]] MinMaxScaler() Roc_AUC validation set: 0.9091666666666667 ROC_AUC test set: 0.8628237259816207 Accuracy validation set:0.844559585492228 Accuracy test set:0.855072463768116 Confusion Matrix: [[ 94 29] [ 11 142]] RobustScaler() Roc_AUC validation set: 0.9087499999999998 ROC_AUC test set: 0.8030830166461489 Accuracy validation set:0.8238341968911918 Accuracy test set:0.7898550724637681 Confusion Matrix: [[111 12] [ 46 107]] PowerTransformer()
C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\preprocessing\_data.py:3253: RuntimeWarning: divide by zero encountered in log loglike = -n_samples / 2 * np.log(x_trans.var())
Roc_AUC validation set: 0.8934166666666666 ROC_AUC test set: 0.8550233833252701 Accuracy validation set:0.8290155440414507 Accuracy test set:0.855072463768116 Confusion Matrix: [[100 23] [ 17 136]]
from sklearn.neighbors import KNeighborsClassifier
number_n=np.linspace(1,50,50).astype(int)
knn_score=[]
for n in number_n:
knn_clf=KNeighborsClassifier(n_neighbors=n)
knn_clf.fit(X_train,y_train)
knn_score.append(roc_auc_score(knn_clf.predict(X_val),y_val))
plt.plot(number_n,knn_score)
plt.xticks(np.linspace(0,50,26))
plt.show()
knn_clf=KNeighborsClassifier(n_neighbors=30)
knn_clf.fit(X_train,y_train)
knn_cv,knn_score,knn_acc_v,knn_acc_t=clf_score(knn_clf,X_val,y_val,X_test,y_test,cv=20)
Roc_AUC validation set: 0.9017916666666667 ROC_AUC test set: 0.8537456678219141 Accuracy validation set:0.844559585492228 Accuracy test set:0.855072463768116 Confusion Matrix: [[102 21] [ 19 134]]
knn_famd_acc,knn_pca_acc,knn_kpca_acc=plot_perf_comp('KNN',knn_clf,knn_cv.mean(),knn_score,knn_acc_v,knn_acc_t)
transformation_comp('KNN',knn_clf)
StandardScaler() Roc_AUC validation set: 0.9017916666666667 ROC_AUC test set: 0.8537456678219141 Accuracy validation set:0.844559585492228 Accuracy test set:0.855072463768116 Confusion Matrix: [[102 21] [ 19 134]] MinMaxScaler() Roc_AUC validation set: 0.8994166666666666 ROC_AUC test set: 0.8701533406352683 Accuracy validation set:0.844559585492228 Accuracy test set:0.8659420289855072 Confusion Matrix: [[ 98 25] [ 12 141]] RobustScaler() Roc_AUC validation set: 0.9290833333333334 ROC_AUC test set: 0.8274231678486997 Accuracy validation set:0.8393782383419689 Accuracy test set:0.8260869565217391 Confusion Matrix: [[108 15] [ 33 120]] PowerTransformer()
C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\preprocessing\_data.py:3253: RuntimeWarning: divide by zero encountered in log loglike = -n_samples / 2 * np.log(x_trans.var())
Roc_AUC validation set: 0.9043333333333333 ROC_AUC test set: 0.8583458485303583 Accuracy validation set:0.8549222797927462 Accuracy test set:0.8586956521739131 Confusion Matrix: [[101 22] [ 17 136]]
from sklearn.naive_bayes import GaussianNB
nb_clf=GaussianNB()
param_grid_nb={'var_smoothing': np.logspace(-1,1, num=100)}
nb_bp=GridSearchCV(nb_clf,param_grid=param_grid_nb,cv=20)
nb_bp.fit(X_train,y_train)
print(nb_bp.best_params_)
nb_cv,nb_score,nb_acc_v,nb_acc_t=clf_score(nb_bp.best_estimator_,X_val,y_val,X_test,y_test,cv=20)
{'var_smoothing': 2.4770763559917115}
Roc_AUC validation set: 0.9194166666666665
ROC_AUC test set: 0.8571960826059186
Accuracy validation set:0.8238341968911918
Accuracy test set:0.8586956521739131
Confusion Matrix:
[[103 20]
[ 19 134]]
nb_famd_acc,nb_pca_acc,nb_kpca_acc=plot_perf_comp('Gaussian Naive Bayes',nb_bp.best_estimator_,nb_cv.mean(),nb_score,nb_acc_v,nb_acc_t)
transformation_comp('Gaussian Naive Bayes',nb_bp.best_estimator_)
StandardScaler() Roc_AUC validation set: 0.9194166666666665 ROC_AUC test set: 0.8571960826059186 Accuracy validation set:0.8238341968911918 Accuracy test set:0.8586956521739131 Confusion Matrix: [[103 20] [ 19 134]] MinMaxScaler() Roc_AUC validation set: 0.90525 ROC_AUC test set: 0.877710843373494 Accuracy validation set:0.8341968911917098 Accuracy test set:0.8731884057971014 Confusion Matrix: [[ 99 24] [ 11 142]] RobustScaler() Roc_AUC validation set: 0.9315 ROC_AUC test set: 0.8369565217391305 Accuracy validation set:0.8238341968911918 Accuracy test set:0.8369565217391305 Confusion Matrix: [[108 15] [ 30 123]] PowerTransformer() Roc_AUC validation set: 0.91225
C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\preprocessing\_data.py:3253: RuntimeWarning: divide by zero encountered in log loglike = -n_samples / 2 * np.log(x_trans.var())
ROC_AUC test set: 0.8571960826059186 Accuracy validation set:0.8186528497409327 Accuracy test set:0.8586956521739131 Confusion Matrix: [[103 20] [ 19 134]]
random_rf_clf=RandomForestClassifier(n_jobs=-1)
random_grid = {'n_estimators': np.linspace(10,1000,331,dtype=int),
'max_features': ['auto', 'sqrt'],
'max_depth': np.linspace(10,255,50,dtype=int),
'min_samples_split': [2,5,10,50],
'min_samples_leaf': [2,4,10,20],
'bootstrap': [True,False]
}
random_rf=RandomizedSearchCV(estimator = random_rf_clf, param_distributions = random_grid, n_iter = 1000, cv = 5, verbose=2)
random_rf.fit(X_val,y_val)
Fitting 5 folds for each of 1000 candidates, totalling 5000 fits [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=277; total time= 0.4s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=277; total time= 0.3s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=277; total time= 0.3s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=277; total time= 0.3s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=277; total time= 0.3s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=901; total time= 0.9s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=901; total time= 1.0s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=901; total time= 1.0s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=901; total time= 0.9s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=901; total time= 0.9s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=910; total time= 1.0s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=910; total time= 1.0s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=910; total time= 1.0s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=910; total time= 1.0s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=910; total time= 1.1s [CV] END bootstrap=False, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=619; total time= 0.8s [CV] END bootstrap=False, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=619; total time= 0.8s [CV] END bootstrap=False, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=619; total time= 0.7s [CV] END bootstrap=False, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=619; total time= 0.8s [CV] END bootstrap=False, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=619; total time= 0.8s [CV] END bootstrap=False, max_depth=70, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=886; total time= 1.1s [CV] END bootstrap=False, max_depth=70, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=886; total time= 1.0s [CV] END bootstrap=False, max_depth=70, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=886; total time= 1.0s [CV] END bootstrap=False, max_depth=70, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=886; total time= 1.0s [CV] END bootstrap=False, max_depth=70, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=886; total time= 1.0s [CV] END bootstrap=False, max_depth=170, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=73; total time= 0.1s [CV] END bootstrap=False, max_depth=170, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=73; total time= 0.1s [CV] END bootstrap=False, max_depth=170, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=73; total time= 0.1s [CV] END bootstrap=False, max_depth=170, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=73; total time= 0.1s [CV] END bootstrap=False, max_depth=170, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=73; total time= 0.1s [CV] END bootstrap=False, max_depth=110, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=328; total time= 0.4s [CV] END bootstrap=False, max_depth=110, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=328; total time= 0.5s [CV] END bootstrap=False, max_depth=110, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=328; total time= 0.4s [CV] END bootstrap=False, max_depth=110, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=328; total time= 0.5s [CV] END bootstrap=False, max_depth=110, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=328; total time= 0.4s [CV] END bootstrap=True, max_depth=150, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=874; total time= 1.2s [CV] END bootstrap=True, max_depth=150, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=874; total time= 1.0s [CV] END bootstrap=True, max_depth=150, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=874; total time= 0.8s [CV] END bootstrap=True, max_depth=150, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=874; total time= 1.0s [CV] END bootstrap=True, max_depth=150, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=874; total time= 0.9s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=103; total time= 0.1s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=103; total time= 0.1s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=103; total time= 0.2s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=103; total time= 0.1s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=103; total time= 0.1s [CV] END bootstrap=True, max_depth=15, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=412; total time= 0.4s [CV] END bootstrap=True, max_depth=15, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=412; total time= 0.4s [CV] END bootstrap=True, max_depth=15, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=412; total time= 0.4s [CV] END bootstrap=True, max_depth=15, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=412; total time= 0.4s [CV] END bootstrap=True, max_depth=15, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=412; total time= 0.4s [CV] END bootstrap=False, max_depth=245, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=988; total time= 1.1s [CV] END bootstrap=False, max_depth=245, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=988; total time= 0.9s [CV] END bootstrap=False, max_depth=245, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=988; total time= 1.0s [CV] END bootstrap=False, max_depth=245, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=988; total time= 1.2s [CV] END bootstrap=False, max_depth=245, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=988; total time= 1.1s [CV] END bootstrap=True, max_depth=45, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=424; total time= 0.5s [CV] END bootstrap=True, max_depth=45, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=424; total time= 0.4s [CV] END bootstrap=True, max_depth=45, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=424; total time= 0.5s [CV] END bootstrap=True, max_depth=45, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=424; total time= 0.5s [CV] END bootstrap=True, max_depth=45, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=424; total time= 0.5s [CV] END bootstrap=True, max_depth=135, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=781; total time= 0.6s [CV] END bootstrap=True, max_depth=135, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=781; total time= 0.7s [CV] END bootstrap=True, max_depth=135, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=781; total time= 0.7s [CV] END bootstrap=True, max_depth=135, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=781; total time= 0.7s [CV] END bootstrap=True, max_depth=135, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=781; total time= 0.8s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=361; total time= 0.6s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=361; total time= 0.5s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=361; total time= 0.6s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=361; total time= 0.6s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=361; total time= 0.5s [CV] END bootstrap=False, max_depth=10, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=220; total time= 0.3s [CV] END bootstrap=False, max_depth=10, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=220; total time= 0.2s [CV] END bootstrap=False, max_depth=10, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=220; total time= 0.3s [CV] END bootstrap=False, max_depth=10, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=220; total time= 0.2s [CV] END bootstrap=False, max_depth=10, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=220; total time= 0.3s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=409; total time= 0.4s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=409; total time= 0.4s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=409; total time= 0.5s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=409; total time= 0.4s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=409; total time= 0.5s [CV] END bootstrap=True, max_depth=175, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=523; total time= 0.4s [CV] END bootstrap=True, max_depth=175, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=523; total time= 0.5s [CV] END bootstrap=True, max_depth=175, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=523; total time= 0.4s [CV] END bootstrap=True, max_depth=175, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=523; total time= 0.5s [CV] END bootstrap=True, max_depth=175, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=523; total time= 0.4s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=502; total time= 0.5s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=502; total time= 0.7s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=502; total time= 0.4s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=502; total time= 0.5s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=502; total time= 0.7s [CV] END bootstrap=False, max_depth=10, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=163; total time= 0.2s [CV] END bootstrap=False, max_depth=10, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=163; total time= 0.2s [CV] END bootstrap=False, max_depth=10, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=163; total time= 0.3s [CV] END bootstrap=False, max_depth=10, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=163; total time= 0.2s [CV] END bootstrap=False, max_depth=10, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=163; total time= 0.2s [CV] END bootstrap=False, max_depth=45, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=574; total time= 0.7s [CV] END bootstrap=False, max_depth=45, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=574; total time= 0.7s [CV] END bootstrap=False, max_depth=45, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=574; total time= 0.6s [CV] END bootstrap=False, max_depth=45, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=574; total time= 0.6s [CV] END bootstrap=False, max_depth=45, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=574; total time= 0.7s [CV] END bootstrap=True, max_depth=15, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=496; total time= 0.5s [CV] END bootstrap=True, max_depth=15, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=496; total time= 0.6s [CV] END bootstrap=True, max_depth=15, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=496; total time= 0.4s [CV] END bootstrap=True, max_depth=15, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=496; total time= 0.3s [CV] END bootstrap=True, max_depth=15, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=496; total time= 0.5s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=724; total time= 0.8s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=724; total time= 0.8s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=724; total time= 0.8s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=724; total time= 0.8s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=724; total time= 0.8s [CV] END bootstrap=False, max_depth=185, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=958; total time= 1.0s [CV] END bootstrap=False, max_depth=185, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=958; total time= 1.0s [CV] END bootstrap=False, max_depth=185, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=958; total time= 1.0s [CV] END bootstrap=False, max_depth=185, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=958; total time= 0.9s [CV] END bootstrap=False, max_depth=185, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=958; total time= 1.1s [CV] END bootstrap=True, max_depth=135, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=112; total time= 0.1s [CV] END bootstrap=True, max_depth=135, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=112; total time= 0.1s [CV] END bootstrap=True, max_depth=135, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=112; total time= 0.1s [CV] END bootstrap=True, max_depth=135, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=112; total time= 0.1s [CV] END bootstrap=True, max_depth=135, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=112; total time= 0.1s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=544; total time= 0.5s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=544; total time= 0.5s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=544; total time= 0.6s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=544; total time= 0.5s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=544; total time= 0.5s [CV] END bootstrap=True, max_depth=25, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=34; total time= 0.0s [CV] END bootstrap=True, max_depth=25, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=34; total time= 0.0s [CV] END bootstrap=True, max_depth=25, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=34; total time= 0.0s [CV] END bootstrap=True, max_depth=25, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=34; total time= 0.0s [CV] END bootstrap=True, max_depth=25, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=34; total time= 0.0s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=916; total time= 0.9s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=916; total time= 1.0s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=916; total time= 0.9s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=916; total time= 0.9s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=916; total time= 0.8s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=691; total time= 0.9s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=691; total time= 0.7s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=691; total time= 0.8s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=691; total time= 0.7s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=691; total time= 0.8s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=607; total time= 0.8s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=607; total time= 0.7s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=607; total time= 0.7s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=607; total time= 0.7s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=607; total time= 0.7s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=373; total time= 0.4s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=373; total time= 0.8s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=373; total time= 0.3s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=373; total time= 0.3s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=373; total time= 0.3s [CV] END bootstrap=True, max_depth=45, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=343; total time= 0.5s [CV] END bootstrap=True, max_depth=45, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=343; total time= 0.5s [CV] END bootstrap=True, max_depth=45, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=343; total time= 0.4s [CV] END bootstrap=True, max_depth=45, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=343; total time= 0.4s [CV] END bootstrap=True, max_depth=45, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=343; total time= 0.5s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=334; total time= 0.4s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=334; total time= 0.4s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=334; total time= 0.4s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=334; total time= 0.4s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=334; total time= 0.4s [CV] END bootstrap=False, max_depth=30, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=364; total time= 0.5s [CV] END bootstrap=False, max_depth=30, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=364; total time= 0.5s [CV] END bootstrap=False, max_depth=30, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=364; total time= 0.5s [CV] END bootstrap=False, max_depth=30, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=364; total time= 0.5s [CV] END bootstrap=False, max_depth=30, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=364; total time= 0.5s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=940; total time= 1.1s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=940; total time= 1.0s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=940; total time= 1.0s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=940; total time= 1.0s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=940; total time= 0.9s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=541; total time= 0.5s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=541; total time= 0.5s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=541; total time= 0.6s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=541; total time= 0.5s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=541; total time= 0.5s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=391; total time= 0.3s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=391; total time= 0.4s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=391; total time= 0.3s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=391; total time= 0.4s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=391; total time= 0.4s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=772; total time= 0.6s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=772; total time= 0.6s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=772; total time= 0.6s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=772; total time= 0.6s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=772; total time= 0.6s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=793; total time= 0.8s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=793; total time= 0.7s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=793; total time= 0.8s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=793; total time= 0.7s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=793; total time= 0.8s [CV] END bootstrap=False, max_depth=230, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=193; total time= 0.2s [CV] END bootstrap=False, max_depth=230, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=193; total time= 0.3s [CV] END bootstrap=False, max_depth=230, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=193; total time= 0.2s [CV] END bootstrap=False, max_depth=230, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=193; total time= 0.2s [CV] END bootstrap=False, max_depth=230, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=193; total time= 0.2s [CV] END bootstrap=True, max_depth=140, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=475; total time= 0.7s [CV] END bootstrap=True, max_depth=140, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=475; total time= 0.6s [CV] END bootstrap=True, max_depth=140, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=475; total time= 0.6s [CV] END bootstrap=True, max_depth=140, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=475; total time= 0.5s [CV] END bootstrap=True, max_depth=140, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=475; total time= 0.7s [CV] END bootstrap=False, max_depth=140, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=553; total time= 0.6s [CV] END bootstrap=False, max_depth=140, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=553; total time= 0.5s [CV] END bootstrap=False, max_depth=140, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=553; total time= 0.6s [CV] END bootstrap=False, max_depth=140, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=553; total time= 0.5s [CV] END bootstrap=False, max_depth=140, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=553; total time= 0.5s [CV] END bootstrap=False, max_depth=175, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=829; total time= 0.8s [CV] END bootstrap=False, max_depth=175, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=829; total time= 0.8s [CV] END bootstrap=False, max_depth=175, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=829; total time= 0.9s [CV] END bootstrap=False, max_depth=175, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=829; total time= 1.0s [CV] END bootstrap=False, max_depth=175, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=829; total time= 0.8s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=412; total time= 0.4s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=412; total time= 0.4s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=412; total time= 0.5s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=412; total time= 0.4s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=412; total time= 0.5s [CV] END bootstrap=False, max_depth=225, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=208; total time= 0.3s [CV] END bootstrap=False, max_depth=225, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=208; total time= 0.2s [CV] END bootstrap=False, max_depth=225, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=208; total time= 0.2s [CV] END bootstrap=False, max_depth=225, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=208; total time= 0.2s [CV] END bootstrap=False, max_depth=225, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=208; total time= 0.2s [CV] END bootstrap=False, max_depth=200, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=670; total time= 0.8s [CV] END bootstrap=False, max_depth=200, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=670; total time= 0.8s [CV] END bootstrap=False, max_depth=200, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=670; total time= 0.7s [CV] END bootstrap=False, max_depth=200, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=670; total time= 0.7s [CV] END bootstrap=False, max_depth=200, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=670; total time= 0.5s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=817; total time= 0.8s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=817; total time= 0.8s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=817; total time= 0.8s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=817; total time= 0.8s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=817; total time= 0.8s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=721; total time= 0.8s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=721; total time= 0.9s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=721; total time= 0.7s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=721; total time= 0.8s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=721; total time= 0.8s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=556; total time= 0.6s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=556; total time= 0.6s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=556; total time= 0.5s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=556; total time= 0.6s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=556; total time= 0.6s [CV] END bootstrap=True, max_depth=220, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=481; total time= 0.6s [CV] END bootstrap=True, max_depth=220, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=481; total time= 0.6s [CV] END bootstrap=True, max_depth=220, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=481; total time= 0.6s [CV] END bootstrap=True, max_depth=220, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=481; total time= 0.6s [CV] END bootstrap=True, max_depth=220, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=481; total time= 0.6s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=775; total time= 0.7s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=775; total time= 0.7s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=775; total time= 0.6s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=775; total time= 0.7s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=775; total time= 0.7s [CV] END bootstrap=True, max_depth=200, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=520; total time= 0.4s [CV] END bootstrap=True, max_depth=200, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=520; total time= 0.5s [CV] END bootstrap=True, max_depth=200, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=520; total time= 0.5s [CV] END bootstrap=True, max_depth=200, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=520; total time= 0.6s [CV] END bootstrap=True, max_depth=200, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=520; total time= 0.5s [CV] END bootstrap=False, max_depth=90, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=589; total time= 0.8s [CV] END bootstrap=False, max_depth=90, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=589; total time= 0.7s [CV] END bootstrap=False, max_depth=90, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=589; total time= 0.6s [CV] END bootstrap=False, max_depth=90, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=589; total time= 0.7s [CV] END bootstrap=False, max_depth=90, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=589; total time= 0.7s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=682; total time= 0.8s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=682; total time= 0.7s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=682; total time= 0.8s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=682; total time= 0.7s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=682; total time= 0.8s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=754; total time= 0.6s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=754; total time= 0.5s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=754; total time= 0.8s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=754; total time= 0.6s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=754; total time= 0.5s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=361; total time= 0.6s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=361; total time= 0.5s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=361; total time= 0.5s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=361; total time= 0.5s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=361; total time= 0.5s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=280; total time= 0.3s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=280; total time= 0.3s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=280; total time= 0.3s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=280; total time= 0.3s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=280; total time= 0.3s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=181; total time= 0.3s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=181; total time= 0.3s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=181; total time= 0.1s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=181; total time= 0.2s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=181; total time= 0.3s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=634; total time= 0.8s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=634; total time= 0.8s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=634; total time= 0.8s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=634; total time= 0.8s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=634; total time= 0.9s [CV] END bootstrap=False, max_depth=55, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=313; total time= 0.5s [CV] END bootstrap=False, max_depth=55, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=313; total time= 0.3s [CV] END bootstrap=False, max_depth=55, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=313; total time= 0.4s [CV] END bootstrap=False, max_depth=55, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=313; total time= 0.4s [CV] END bootstrap=False, max_depth=55, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=313; total time= 0.4s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=286; total time= 0.3s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=286; total time= 0.3s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=286; total time= 0.3s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=286; total time= 0.3s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=286; total time= 0.3s [CV] END bootstrap=False, max_depth=210, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=718; total time= 0.8s [CV] END bootstrap=False, max_depth=210, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=718; total time= 0.8s [CV] END bootstrap=False, max_depth=210, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=718; total time= 0.8s [CV] END bootstrap=False, max_depth=210, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=718; total time= 0.7s [CV] END bootstrap=False, max_depth=210, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=718; total time= 0.9s [CV] END bootstrap=False, max_depth=155, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=262; total time= 0.2s [CV] END bootstrap=False, max_depth=155, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=262; total time= 0.2s [CV] END bootstrap=False, max_depth=155, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=262; total time= 0.2s [CV] END bootstrap=False, max_depth=155, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=262; total time= 0.2s [CV] END bootstrap=False, max_depth=155, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=262; total time= 0.3s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=694; total time= 0.9s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=694; total time= 0.8s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=694; total time= 0.7s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=694; total time= 0.7s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=694; total time= 0.7s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=709; total time= 0.8s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=709; total time= 0.9s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=709; total time= 0.9s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=709; total time= 0.8s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=709; total time= 0.8s [CV] END bootstrap=False, max_depth=40, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=478; total time= 0.5s [CV] END bootstrap=False, max_depth=40, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=478; total time= 0.6s [CV] END bootstrap=False, max_depth=40, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=478; total time= 0.6s [CV] END bootstrap=False, max_depth=40, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=478; total time= 0.6s [CV] END bootstrap=False, max_depth=40, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=478; total time= 0.6s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=139; total time= 0.2s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=139; total time= 0.2s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=139; total time= 0.2s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=139; total time= 0.2s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=139; total time= 0.2s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=427; total time= 0.5s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=427; total time= 0.5s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=427; total time= 0.4s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=427; total time= 0.4s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=427; total time= 0.5s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=448; total time= 0.5s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=448; total time= 0.5s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=448; total time= 0.6s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=448; total time= 0.5s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=448; total time= 0.5s [CV] END bootstrap=True, max_depth=245, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=931; total time= 0.9s [CV] END bootstrap=True, max_depth=245, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=931; total time= 1.0s [CV] END bootstrap=True, max_depth=245, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=931; total time= 1.0s [CV] END bootstrap=True, max_depth=245, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=931; total time= 0.9s [CV] END bootstrap=True, max_depth=245, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=931; total time= 1.0s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=922; total time= 0.9s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=922; total time= 0.9s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=922; total time= 0.9s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=922; total time= 0.9s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=922; total time= 0.9s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=694; total time= 0.7s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=694; total time= 0.7s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=694; total time= 0.8s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=694; total time= 0.8s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=694; total time= 0.8s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=517; total time= 0.7s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=517; total time= 0.4s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=517; total time= 0.4s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=517; total time= 0.5s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=517; total time= 0.4s [CV] END bootstrap=False, max_depth=10, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=220; total time= 0.2s [CV] END bootstrap=False, max_depth=10, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=220; total time= 0.2s [CV] END bootstrap=False, max_depth=10, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=220; total time= 0.2s [CV] END bootstrap=False, max_depth=10, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=220; total time= 0.3s [CV] END bootstrap=False, max_depth=10, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=220; total time= 0.3s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=577; total time= 0.6s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=577; total time= 0.6s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=577; total time= 0.7s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=577; total time= 0.7s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=577; total time= 0.6s [CV] END bootstrap=True, max_depth=60, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=997; total time= 0.8s [CV] END bootstrap=True, max_depth=60, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=997; total time= 1.0s [CV] END bootstrap=True, max_depth=60, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=997; total time= 0.9s [CV] END bootstrap=True, max_depth=60, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=997; total time= 1.1s [CV] END bootstrap=True, max_depth=60, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=997; total time= 1.1s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=676; total time= 0.7s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=676; total time= 0.7s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=676; total time= 0.7s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=676; total time= 0.7s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=676; total time= 0.7s [CV] END bootstrap=True, max_depth=70, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=979; total time= 1.1s [CV] END bootstrap=True, max_depth=70, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=979; total time= 1.2s [CV] END bootstrap=True, max_depth=70, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=979; total time= 1.0s [CV] END bootstrap=True, max_depth=70, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=979; total time= 1.0s [CV] END bootstrap=True, max_depth=70, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=979; total time= 1.0s [CV] END bootstrap=True, max_depth=50, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=85; total time= 0.1s [CV] END bootstrap=True, max_depth=50, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=85; total time= 0.1s [CV] END bootstrap=True, max_depth=50, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=85; total time= 0.1s [CV] END bootstrap=True, max_depth=50, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=85; total time= 0.1s [CV] END bootstrap=True, max_depth=50, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=85; total time= 0.1s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=64; total time= 0.1s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=64; total time= 0.0s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=64; total time= 0.1s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=64; total time= 0.1s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=64; total time= 0.1s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=115; total time= 0.2s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=115; total time= 0.1s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=115; total time= 0.2s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=115; total time= 0.1s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=115; total time= 0.1s [CV] END bootstrap=False, max_depth=190, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=403; total time= 0.5s [CV] END bootstrap=False, max_depth=190, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=403; total time= 0.5s [CV] END bootstrap=False, max_depth=190, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=403; total time= 0.4s [CV] END bootstrap=False, max_depth=190, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=403; total time= 0.4s [CV] END bootstrap=False, max_depth=190, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=403; total time= 0.4s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=565; total time= 0.6s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=565; total time= 0.7s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=565; total time= 0.6s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=565; total time= 0.6s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=565; total time= 0.6s [CV] END bootstrap=True, max_depth=40, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=100; total time= 0.1s [CV] END bootstrap=True, max_depth=40, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=100; total time= 0.1s [CV] END bootstrap=True, max_depth=40, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=100; total time= 0.1s [CV] END bootstrap=True, max_depth=40, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=100; total time= 0.1s [CV] END bootstrap=True, max_depth=40, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=100; total time= 0.1s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=145; total time= 0.2s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=145; total time= 0.2s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=145; total time= 0.2s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=145; total time= 0.2s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=145; total time= 0.2s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=166; total time= 0.2s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=166; total time= 0.2s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=166; total time= 0.3s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=166; total time= 0.2s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=166; total time= 0.2s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=280; total time= 0.3s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=280; total time= 0.3s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=280; total time= 0.3s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=280; total time= 0.3s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=280; total time= 0.3s [CV] END bootstrap=False, max_depth=60, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=862; total time= 0.9s [CV] END bootstrap=False, max_depth=60, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=862; total time= 0.9s [CV] END bootstrap=False, max_depth=60, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=862; total time= 1.0s [CV] END bootstrap=False, max_depth=60, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=862; total time= 0.8s [CV] END bootstrap=False, max_depth=60, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=862; total time= 0.9s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=973; total time= 0.9s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=973; total time= 0.9s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=973; total time= 1.0s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=973; total time= 1.1s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=973; total time= 1.0s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=940; total time= 1.0s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=940; total time= 0.9s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=940; total time= 1.0s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=940; total time= 0.9s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=940; total time= 0.9s [CV] END bootstrap=False, max_depth=190, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=865; total time= 1.0s [CV] END bootstrap=False, max_depth=190, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=865; total time= 0.8s [CV] END bootstrap=False, max_depth=190, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=865; total time= 1.0s [CV] END bootstrap=False, max_depth=190, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=865; total time= 1.1s [CV] END bootstrap=False, max_depth=190, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=865; total time= 1.0s [CV] END bootstrap=True, max_depth=50, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=541; total time= 0.6s [CV] END bootstrap=True, max_depth=50, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=541; total time= 0.5s [CV] END bootstrap=True, max_depth=50, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=541; total time= 0.5s [CV] END bootstrap=True, max_depth=50, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=541; total time= 0.5s [CV] END bootstrap=True, max_depth=50, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=541; total time= 0.5s [CV] END bootstrap=False, max_depth=90, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=727; total time= 0.7s [CV] END bootstrap=False, max_depth=90, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=727; total time= 0.7s [CV] END bootstrap=False, max_depth=90, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=727; total time= 0.7s [CV] END bootstrap=False, max_depth=90, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=727; total time= 0.8s [CV] END bootstrap=False, max_depth=90, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=727; total time= 0.9s [CV] END bootstrap=True, max_depth=225, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=295; total time= 0.3s [CV] END bootstrap=True, max_depth=225, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=295; total time= 0.5s [CV] END bootstrap=True, max_depth=225, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=295; total time= 0.3s [CV] END bootstrap=True, max_depth=225, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=295; total time= 0.4s [CV] END bootstrap=True, max_depth=225, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=295; total time= 0.4s [CV] END bootstrap=False, max_depth=75, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=106; total time= 0.1s [CV] END bootstrap=False, max_depth=75, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=106; total time= 0.2s [CV] END bootstrap=False, max_depth=75, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=106; total time= 0.1s [CV] END bootstrap=False, max_depth=75, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=106; total time= 0.1s [CV] END bootstrap=False, max_depth=75, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=106; total time= 0.2s [CV] END bootstrap=True, max_depth=25, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=661; total time= 0.7s [CV] END bootstrap=True, max_depth=25, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=661; total time= 0.7s [CV] END bootstrap=True, max_depth=25, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=661; total time= 0.8s [CV] END bootstrap=True, max_depth=25, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=661; total time= 0.7s [CV] END bootstrap=True, max_depth=25, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=661; total time= 0.6s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=493; total time= 0.7s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=493; total time= 0.6s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=493; total time= 0.6s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=493; total time= 0.7s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=493; total time= 0.7s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=298; total time= 0.3s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=298; total time= 0.3s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=298; total time= 0.3s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=298; total time= 0.4s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=298; total time= 0.3s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=328; total time= 0.5s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=328; total time= 0.5s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=328; total time= 0.5s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=328; total time= 0.4s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=328; total time= 0.4s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=745; total time= 0.8s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=745; total time= 0.8s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=745; total time= 0.7s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=745; total time= 0.8s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=745; total time= 0.8s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=349; total time= 0.5s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=349; total time= 0.5s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=349; total time= 0.5s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=349; total time= 0.5s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=349; total time= 0.5s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=586; total time= 0.6s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=586; total time= 0.6s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=586; total time= 0.6s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=586; total time= 0.6s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=586; total time= 0.7s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=871; total time= 1.1s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=871; total time= 1.0s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=871; total time= 0.8s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=871; total time= 1.0s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=871; total time= 1.0s [CV] END bootstrap=True, max_depth=15, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=304; total time= 0.4s [CV] END bootstrap=True, max_depth=15, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=304; total time= 0.4s [CV] END bootstrap=True, max_depth=15, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=304; total time= 0.3s [CV] END bootstrap=True, max_depth=15, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=304; total time= 0.3s [CV] END bootstrap=True, max_depth=15, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=304; total time= 0.4s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=127; total time= 0.1s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=127; total time= 0.1s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=127; total time= 0.1s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=127; total time= 0.1s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=127; total time= 0.2s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=487; total time= 0.6s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=487; total time= 0.6s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=487; total time= 0.7s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=487; total time= 0.6s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=487; total time= 0.5s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=475; total time= 0.6s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=475; total time= 0.6s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=475; total time= 0.5s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=475; total time= 0.6s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=475; total time= 0.7s [CV] END bootstrap=False, max_depth=35, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=724; total time= 0.9s [CV] END bootstrap=False, max_depth=35, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=724; total time= 0.8s [CV] END bootstrap=False, max_depth=35, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=724; total time= 0.8s [CV] END bootstrap=False, max_depth=35, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=724; total time= 0.7s [CV] END bootstrap=False, max_depth=35, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=724; total time= 0.8s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=658; total time= 0.7s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=658; total time= 0.7s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=658; total time= 0.8s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=658; total time= 0.9s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=658; total time= 0.7s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=994; total time= 1.0s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=994; total time= 1.1s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=994; total time= 0.9s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=994; total time= 1.0s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=994; total time= 0.9s [CV] END bootstrap=False, max_depth=140, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=673; total time= 0.7s [CV] END bootstrap=False, max_depth=140, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=673; total time= 0.7s [CV] END bootstrap=False, max_depth=140, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=673; total time= 0.8s [CV] END bootstrap=False, max_depth=140, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=673; total time= 0.7s [CV] END bootstrap=False, max_depth=140, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=673; total time= 0.7s [CV] END bootstrap=True, max_depth=180, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=685; total time= 0.7s [CV] END bootstrap=True, max_depth=180, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=685; total time= 0.8s [CV] END bootstrap=True, max_depth=180, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=685; total time= 0.8s [CV] END bootstrap=True, max_depth=180, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=685; total time= 0.7s [CV] END bootstrap=True, max_depth=180, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=685; total time= 0.8s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=643; total time= 0.8s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=643; total time= 0.8s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=643; total time= 0.8s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=643; total time= 0.9s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=643; total time= 0.8s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=676; total time= 0.6s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=676; total time= 0.7s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=676; total time= 0.7s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=676; total time= 0.7s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=676; total time= 0.8s [CV] END bootstrap=False, max_depth=175, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=559; total time= 0.6s [CV] END bootstrap=False, max_depth=175, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=559; total time= 0.5s [CV] END bootstrap=False, max_depth=175, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=559; total time= 0.6s [CV] END bootstrap=False, max_depth=175, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=559; total time= 0.6s [CV] END bootstrap=False, max_depth=175, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=559; total time= 0.5s [CV] END bootstrap=False, max_depth=45, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=817; total time= 0.9s [CV] END bootstrap=False, max_depth=45, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=817; total time= 0.8s [CV] END bootstrap=False, max_depth=45, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=817; total time= 1.0s [CV] END bootstrap=False, max_depth=45, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=817; total time= 0.9s [CV] END bootstrap=False, max_depth=45, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=817; total time= 0.8s [CV] END bootstrap=True, max_depth=40, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=313; total time= 0.4s [CV] END bootstrap=True, max_depth=40, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=313; total time= 0.4s [CV] END bootstrap=True, max_depth=40, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=313; total time= 0.4s [CV] END bootstrap=True, max_depth=40, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=313; total time= 0.3s [CV] END bootstrap=True, max_depth=40, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=313; total time= 0.4s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=958; total time= 1.0s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=958; total time= 0.9s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=958; total time= 1.0s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=958; total time= 1.0s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=958; total time= 1.0s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=253; total time= 0.2s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=253; total time= 0.2s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=253; total time= 0.2s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=253; total time= 0.3s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=253; total time= 0.3s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=613; total time= 0.9s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=613; total time= 0.7s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=613; total time= 0.7s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=613; total time= 0.8s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=613; total time= 0.8s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=553; total time= 0.5s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=553; total time= 0.5s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=553; total time= 0.5s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=553; total time= 0.6s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=553; total time= 0.4s [CV] END bootstrap=False, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=874; total time= 0.9s [CV] END bootstrap=False, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=874; total time= 0.9s [CV] END bootstrap=False, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=874; total time= 1.0s [CV] END bootstrap=False, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=874; total time= 1.0s [CV] END bootstrap=False, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=874; total time= 1.0s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=76; total time= 0.1s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=76; total time= 0.1s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=76; total time= 0.1s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=76; total time= 0.1s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=76; total time= 0.1s [CV] END bootstrap=True, max_depth=85, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=727; total time= 1.0s [CV] END bootstrap=True, max_depth=85, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=727; total time= 0.8s [CV] END bootstrap=True, max_depth=85, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=727; total time= 0.9s [CV] END bootstrap=True, max_depth=85, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=727; total time= 0.8s [CV] END bootstrap=True, max_depth=85, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=727; total time= 0.9s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=664; total time= 0.8s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=664; total time= 0.6s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=664; total time= 0.7s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=664; total time= 0.7s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=664; total time= 0.7s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=22; total time= 0.0s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=22; total time= 0.0s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=22; total time= 0.0s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=22; total time= 0.0s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=22; total time= 0.0s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=163; total time= 0.2s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=163; total time= 0.2s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=163; total time= 0.3s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=163; total time= 0.2s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=163; total time= 0.3s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=313; total time= 0.3s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=313; total time= 0.4s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=313; total time= 0.4s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=313; total time= 0.3s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=313; total time= 0.3s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=439; total time= 0.4s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=439; total time= 0.5s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=439; total time= 0.5s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=439; total time= 0.5s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=439; total time= 0.5s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=277; total time= 0.3s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=277; total time= 0.3s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=277; total time= 0.3s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=277; total time= 0.3s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=277; total time= 0.4s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=367; total time= 0.5s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=367; total time= 0.6s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=367; total time= 0.5s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=367; total time= 0.6s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=367; total time= 0.5s [CV] END bootstrap=True, max_depth=85, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=859; total time= 0.9s [CV] END bootstrap=True, max_depth=85, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=859; total time= 0.9s [CV] END bootstrap=True, max_depth=85, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=859; total time= 0.9s [CV] END bootstrap=True, max_depth=85, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=859; total time= 0.8s [CV] END bootstrap=True, max_depth=85, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=859; total time= 1.0s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=439; total time= 0.5s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=439; total time= 0.4s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=439; total time= 0.5s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=439; total time= 0.4s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=439; total time= 0.5s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=544; total time= 0.5s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=544; total time= 0.5s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=544; total time= 0.5s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=544; total time= 0.5s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=544; total time= 0.6s [CV] END bootstrap=False, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=124; total time= 0.1s [CV] END bootstrap=False, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=124; total time= 0.1s [CV] END bootstrap=False, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=124; total time= 0.1s [CV] END bootstrap=False, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=124; total time= 0.1s [CV] END bootstrap=False, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=124; total time= 0.2s [CV] END bootstrap=False, max_depth=25, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=250; total time= 0.2s [CV] END bootstrap=False, max_depth=25, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=250; total time= 0.2s [CV] END bootstrap=False, max_depth=25, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=250; total time= 0.2s [CV] END bootstrap=False, max_depth=25, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=250; total time= 0.3s [CV] END bootstrap=False, max_depth=25, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=250; total time= 0.2s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=460; total time= 0.5s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=460; total time= 0.6s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=460; total time= 0.6s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=460; total time= 0.4s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=460; total time= 0.6s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=61; total time= 0.0s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=61; total time= 0.1s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=61; total time= 0.1s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=61; total time= 0.0s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=61; total time= 0.0s [CV] END bootstrap=True, max_depth=80, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=967; total time= 1.0s [CV] END bootstrap=True, max_depth=80, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=967; total time= 0.9s [CV] END bootstrap=True, max_depth=80, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=967; total time= 1.0s [CV] END bootstrap=True, max_depth=80, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=967; total time= 1.0s [CV] END bootstrap=True, max_depth=80, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=967; total time= 1.0s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=424; total time= 0.4s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=424; total time= 0.4s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=424; total time= 0.5s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=424; total time= 0.5s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=424; total time= 0.5s [CV] END bootstrap=False, max_depth=220, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=991; total time= 1.0s [CV] END bootstrap=False, max_depth=220, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=991; total time= 1.0s [CV] END bootstrap=False, max_depth=220, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=991; total time= 1.0s [CV] END bootstrap=False, max_depth=220, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=991; total time= 0.9s [CV] END bootstrap=False, max_depth=220, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=991; total time= 1.1s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=316; total time= 0.3s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=316; total time= 0.5s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=316; total time= 0.4s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=316; total time= 0.4s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=316; total time= 0.4s [CV] END bootstrap=True, max_depth=50, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=913; total time= 0.9s [CV] END bootstrap=True, max_depth=50, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=913; total time= 0.8s [CV] END bootstrap=True, max_depth=50, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=913; total time= 1.1s [CV] END bootstrap=True, max_depth=50, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=913; total time= 0.8s [CV] END bootstrap=True, max_depth=50, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=913; total time= 0.7s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=943; total time= 0.9s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=943; total time= 0.9s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=943; total time= 1.0s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=943; total time= 0.9s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=943; total time= 1.0s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=109; total time= 0.2s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=109; total time= 0.1s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=109; total time= 0.1s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=109; total time= 0.1s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=109; total time= 0.1s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=799; total time= 0.8s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=799; total time= 0.8s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=799; total time= 0.8s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=799; total time= 0.7s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=799; total time= 0.7s [CV] END bootstrap=False, max_depth=225, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=637; total time= 0.7s [CV] END bootstrap=False, max_depth=225, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=637; total time= 0.7s [CV] END bootstrap=False, max_depth=225, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=637; total time= 0.8s [CV] END bootstrap=False, max_depth=225, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=637; total time= 0.8s [CV] END bootstrap=False, max_depth=225, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=637; total time= 0.8s [CV] END bootstrap=False, max_depth=245, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=823; total time= 0.9s [CV] END bootstrap=False, max_depth=245, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=823; total time= 0.8s [CV] END bootstrap=False, max_depth=245, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=823; total time= 0.8s [CV] END bootstrap=False, max_depth=245, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=823; total time= 0.8s [CV] END bootstrap=False, max_depth=245, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=823; total time= 0.8s [CV] END bootstrap=False, max_depth=175, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=814; total time= 0.9s [CV] END bootstrap=False, max_depth=175, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=814; total time= 0.8s [CV] END bootstrap=False, max_depth=175, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=814; total time= 0.8s [CV] END bootstrap=False, max_depth=175, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=814; total time= 0.8s [CV] END bootstrap=False, max_depth=175, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=814; total time= 0.7s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=43; total time= 0.0s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=43; total time= 0.0s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=43; total time= 0.0s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=43; total time= 0.0s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=43; total time= 0.0s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=406; total time= 0.4s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=406; total time= 0.5s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=406; total time= 0.4s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=406; total time= 0.4s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=406; total time= 0.4s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=253; total time= 0.3s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=253; total time= 0.3s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=253; total time= 0.2s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=253; total time= 0.2s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=253; total time= 0.2s [CV] END bootstrap=False, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=181; total time= 0.2s [CV] END bootstrap=False, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=181; total time= 0.2s [CV] END bootstrap=False, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=181; total time= 0.2s [CV] END bootstrap=False, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=181; total time= 0.3s [CV] END bootstrap=False, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=181; total time= 0.2s [CV] END bootstrap=True, max_depth=85, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=91; total time= 0.1s [CV] END bootstrap=True, max_depth=85, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=91; total time= 0.1s [CV] END bootstrap=True, max_depth=85, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=91; total time= 0.1s [CV] END bootstrap=True, max_depth=85, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=91; total time= 0.1s [CV] END bootstrap=True, max_depth=85, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=91; total time= 0.1s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=115; total time= 0.1s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=115; total time= 0.1s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=115; total time= 0.2s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=115; total time= 0.1s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=115; total time= 0.1s [CV] END bootstrap=False, max_depth=110, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=946; total time= 1.0s [CV] END bootstrap=False, max_depth=110, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=946; total time= 1.0s [CV] END bootstrap=False, max_depth=110, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=946; total time= 0.9s [CV] END bootstrap=False, max_depth=110, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=946; total time= 1.0s [CV] END bootstrap=False, max_depth=110, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=946; total time= 1.0s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=424; total time= 0.5s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=424; total time= 0.5s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=424; total time= 0.5s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=424; total time= 0.4s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=424; total time= 0.5s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=433; total time= 0.5s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=433; total time= 0.5s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=433; total time= 0.5s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=433; total time= 0.5s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=433; total time= 0.5s [CV] END bootstrap=True, max_depth=230, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=103; total time= 0.2s [CV] END bootstrap=True, max_depth=230, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=103; total time= 0.1s [CV] END bootstrap=True, max_depth=230, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=103; total time= 0.2s [CV] END bootstrap=True, max_depth=230, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=103; total time= 0.1s [CV] END bootstrap=True, max_depth=230, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=103; total time= 0.1s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=487; total time= 0.7s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=487; total time= 0.6s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=487; total time= 0.7s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=487; total time= 0.8s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=487; total time= 0.6s [CV] END bootstrap=True, max_depth=25, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=334; total time= 0.4s [CV] END bootstrap=True, max_depth=25, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=334; total time= 0.5s [CV] END bootstrap=True, max_depth=25, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=334; total time= 0.4s [CV] END bootstrap=True, max_depth=25, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=334; total time= 0.5s [CV] END bootstrap=True, max_depth=25, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=334; total time= 0.5s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=328; total time= 0.5s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=328; total time= 0.4s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=328; total time= 0.4s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=328; total time= 0.5s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=328; total time= 0.4s [CV] END bootstrap=True, max_depth=230, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=991; total time= 1.2s [CV] END bootstrap=True, max_depth=230, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=991; total time= 1.1s [CV] END bootstrap=True, max_depth=230, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=991; total time= 1.2s [CV] END bootstrap=True, max_depth=230, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=991; total time= 1.0s [CV] END bootstrap=True, max_depth=230, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=991; total time= 1.1s [CV] END bootstrap=True, max_depth=175, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=685; total time= 0.7s [CV] END bootstrap=True, max_depth=175, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=685; total time= 0.7s [CV] END bootstrap=True, max_depth=175, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=685; total time= 0.7s [CV] END bootstrap=True, max_depth=175, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=685; total time= 0.7s [CV] END bootstrap=True, max_depth=175, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=685; total time= 0.7s [CV] END bootstrap=False, max_depth=130, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=997; total time= 1.1s [CV] END bootstrap=False, max_depth=130, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=997; total time= 1.0s [CV] END bootstrap=False, max_depth=130, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=997; total time= 1.1s [CV] END bootstrap=False, max_depth=130, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=997; total time= 1.0s [CV] END bootstrap=False, max_depth=130, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=997; total time= 1.0s [CV] END bootstrap=True, max_depth=245, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=625; total time= 0.8s [CV] END bootstrap=True, max_depth=245, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=625; total time= 0.7s [CV] END bootstrap=True, max_depth=245, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=625; total time= 0.7s [CV] END bootstrap=True, max_depth=245, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=625; total time= 0.8s [CV] END bootstrap=True, max_depth=245, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=625; total time= 0.7s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=616; total time= 0.7s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=616; total time= 0.8s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=616; total time= 0.8s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=616; total time= 0.8s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=616; total time= 0.7s [CV] END bootstrap=False, max_depth=170, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=910; total time= 1.1s [CV] END bootstrap=False, max_depth=170, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=910; total time= 1.1s [CV] END bootstrap=False, max_depth=170, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=910; total time= 1.0s [CV] END bootstrap=False, max_depth=170, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=910; total time= 1.1s [CV] END bootstrap=False, max_depth=170, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=910; total time= 1.1s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=811; total time= 0.8s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=811; total time= 0.7s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=811; total time= 0.8s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=811; total time= 0.8s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=811; total time= 0.9s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=481; total time= 0.6s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=481; total time= 0.7s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=481; total time= 0.6s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=481; total time= 0.6s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=481; total time= 0.7s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=148; total time= 0.2s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=148; total time= 0.1s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=148; total time= 0.2s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=148; total time= 0.2s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=148; total time= 0.2s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=70; total time= 0.1s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=70; total time= 0.1s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=70; total time= 0.1s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=70; total time= 0.1s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=70; total time= 0.1s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=682; total time= 0.9s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=682; total time= 0.8s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=682; total time= 0.8s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=682; total time= 0.8s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=682; total time= 0.8s [CV] END bootstrap=False, max_depth=160, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=220; total time= 0.2s [CV] END bootstrap=False, max_depth=160, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=220; total time= 0.2s [CV] END bootstrap=False, max_depth=160, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=220; total time= 0.3s [CV] END bootstrap=False, max_depth=160, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=220; total time= 0.3s [CV] END bootstrap=False, max_depth=160, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=220; total time= 0.3s [CV] END bootstrap=True, max_depth=40, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=871; total time= 1.0s [CV] END bootstrap=True, max_depth=40, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=871; total time= 0.9s [CV] END bootstrap=True, max_depth=40, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=871; total time= 1.0s [CV] END bootstrap=True, max_depth=40, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=871; total time= 0.9s [CV] END bootstrap=True, max_depth=40, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=871; total time= 0.9s [CV] END bootstrap=False, max_depth=205, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=205; total time= 0.2s [CV] END bootstrap=False, max_depth=205, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=205; total time= 0.2s [CV] END bootstrap=False, max_depth=205, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=205; total time= 0.3s [CV] END bootstrap=False, max_depth=205, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=205; total time= 0.2s [CV] END bootstrap=False, max_depth=205, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=205; total time= 0.3s [CV] END bootstrap=False, max_depth=60, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=82; total time= 0.1s [CV] END bootstrap=False, max_depth=60, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=82; total time= 0.0s [CV] END bootstrap=False, max_depth=60, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=82; total time= 0.0s [CV] END bootstrap=False, max_depth=60, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=82; total time= 0.1s [CV] END bootstrap=False, max_depth=60, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=82; total time= 0.0s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=163; total time= 0.3s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=163; total time= 0.3s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=163; total time= 0.2s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=163; total time= 0.2s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=163; total time= 0.3s [CV] END bootstrap=False, max_depth=210, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=37; total time= 0.0s [CV] END bootstrap=False, max_depth=210, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=37; total time= 0.0s [CV] END bootstrap=False, max_depth=210, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=37; total time= 0.0s [CV] END bootstrap=False, max_depth=210, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=37; total time= 0.0s [CV] END bootstrap=False, max_depth=210, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=37; total time= 0.0s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=649; total time= 0.8s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=649; total time= 0.8s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=649; total time= 0.9s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=649; total time= 0.7s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=649; total time= 0.8s [CV] END bootstrap=False, max_depth=130, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=832; total time= 0.8s [CV] END bootstrap=False, max_depth=130, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=832; total time= 0.9s [CV] END bootstrap=False, max_depth=130, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=832; total time= 0.8s [CV] END bootstrap=False, max_depth=130, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=832; total time= 0.8s [CV] END bootstrap=False, max_depth=130, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=832; total time= 0.8s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=556; total time= 0.5s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=556; total time= 0.5s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=556; total time= 0.5s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=556; total time= 0.6s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=556; total time= 0.5s [CV] END bootstrap=False, max_depth=35, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=775; total time= 0.5s [CV] END bootstrap=False, max_depth=35, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=775; total time= 0.5s [CV] END bootstrap=False, max_depth=35, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=775; total time= 0.5s [CV] END bootstrap=False, max_depth=35, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=775; total time= 0.6s [CV] END bootstrap=False, max_depth=35, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=775; total time= 0.6s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=910; total time= 1.0s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=910; total time= 0.9s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=910; total time= 1.0s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=910; total time= 0.9s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=910; total time= 0.8s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=694; total time= 0.7s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=694; total time= 0.8s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=694; total time= 0.8s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=694; total time= 0.7s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=694; total time= 0.7s [CV] END bootstrap=True, max_depth=30, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=952; total time= 1.0s [CV] END bootstrap=True, max_depth=30, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=952; total time= 1.0s [CV] END bootstrap=True, max_depth=30, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=952; total time= 1.0s [CV] END bootstrap=True, max_depth=30, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=952; total time= 0.9s [CV] END bootstrap=True, max_depth=30, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=952; total time= 0.9s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=367; total time= 0.5s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=367; total time= 0.5s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=367; total time= 0.5s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=367; total time= 0.4s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=367; total time= 0.6s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=430; total time= 0.4s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=430; total time= 0.5s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=430; total time= 0.5s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=430; total time= 0.4s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=430; total time= 0.4s [CV] END bootstrap=True, max_depth=20, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=334; total time= 0.5s [CV] END bootstrap=True, max_depth=20, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=334; total time= 0.5s [CV] END bootstrap=True, max_depth=20, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=334; total time= 0.5s [CV] END bootstrap=True, max_depth=20, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=334; total time= 0.5s [CV] END bootstrap=True, max_depth=20, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=334; total time= 0.4s [CV] END bootstrap=True, max_depth=110, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=733; total time= 0.9s [CV] END bootstrap=True, max_depth=110, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=733; total time= 0.9s [CV] END bootstrap=True, max_depth=110, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=733; total time= 0.9s [CV] END bootstrap=True, max_depth=110, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=733; total time= 1.0s [CV] END bootstrap=True, max_depth=110, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=733; total time= 0.8s [CV] END bootstrap=True, max_depth=140, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=562; total time= 0.7s [CV] END bootstrap=True, max_depth=140, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=562; total time= 0.6s [CV] END bootstrap=True, max_depth=140, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=562; total time= 0.5s [CV] END bootstrap=True, max_depth=140, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=562; total time= 0.6s [CV] END bootstrap=True, max_depth=140, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=562; total time= 0.6s [CV] END bootstrap=True, max_depth=180, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=100; total time= 0.1s [CV] END bootstrap=True, max_depth=180, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=100; total time= 0.1s [CV] END bootstrap=True, max_depth=180, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=100; total time= 0.1s [CV] END bootstrap=True, max_depth=180, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=100; total time= 0.1s [CV] END bootstrap=True, max_depth=180, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=100; total time= 0.1s [CV] END bootstrap=False, max_depth=210, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=253; total time= 0.2s [CV] END bootstrap=False, max_depth=210, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=253; total time= 0.2s [CV] END bootstrap=False, max_depth=210, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=253; total time= 0.3s [CV] END bootstrap=False, max_depth=210, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=253; total time= 0.2s [CV] END bootstrap=False, max_depth=210, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=253; total time= 0.2s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=868; total time= 0.9s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=868; total time= 0.9s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=868; total time= 1.1s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=868; total time= 1.0s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=868; total time= 0.9s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=820; total time= 0.7s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=820; total time= 0.7s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=820; total time= 0.9s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=820; total time= 0.8s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=820; total time= 0.8s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=88; total time= 0.1s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=88; total time= 0.1s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=88; total time= 0.1s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=88; total time= 0.1s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=88; total time= 0.1s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=466; total time= 0.6s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=466; total time= 0.5s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=466; total time= 0.6s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=466; total time= 0.7s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=466; total time= 0.5s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=706; total time= 0.8s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=706; total time= 0.8s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=706; total time= 0.8s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=706; total time= 0.8s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=706; total time= 0.9s [CV] END bootstrap=True, max_depth=115, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=82; total time= 0.1s [CV] END bootstrap=True, max_depth=115, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=82; total time= 0.0s [CV] END bootstrap=True, max_depth=115, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=82; total time= 0.1s [CV] END bootstrap=True, max_depth=115, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=82; total time= 0.0s [CV] END bootstrap=True, max_depth=115, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=82; total time= 0.1s [CV] END bootstrap=True, max_depth=150, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=349; total time= 0.5s [CV] END bootstrap=True, max_depth=150, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=349; total time= 0.5s [CV] END bootstrap=True, max_depth=150, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=349; total time= 0.4s [CV] END bootstrap=True, max_depth=150, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=349; total time= 0.4s [CV] END bootstrap=True, max_depth=150, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=349; total time= 0.5s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=211; total time= 0.2s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=211; total time= 0.2s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=211; total time= 0.3s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=211; total time= 0.2s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=211; total time= 0.2s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=838; total time= 0.8s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=838; total time= 0.8s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=838; total time= 1.0s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=838; total time= 0.8s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=838; total time= 0.9s [CV] END bootstrap=False, max_depth=55, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=487; total time= 0.6s [CV] END bootstrap=False, max_depth=55, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=487; total time= 0.7s [CV] END bootstrap=False, max_depth=55, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=487; total time= 0.8s [CV] END bootstrap=False, max_depth=55, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=487; total time= 0.6s [CV] END bootstrap=False, max_depth=55, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=487; total time= 0.7s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=304; total time= 0.4s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=304; total time= 0.4s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=304; total time= 0.3s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=304; total time= 0.4s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=304; total time= 0.4s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=541; total time= 0.4s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=541; total time= 0.5s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=541; total time= 0.5s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=541; total time= 0.6s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=541; total time= 0.5s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=970; total time= 1.1s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=970; total time= 1.0s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=970; total time= 1.0s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=970; total time= 0.9s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=970; total time= 1.0s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=805; total time= 0.8s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=805; total time= 0.7s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=805; total time= 0.7s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=805; total time= 0.7s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=805; total time= 0.7s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=547; total time= 0.5s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=547; total time= 0.5s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=547; total time= 0.6s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=547; total time= 0.7s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=547; total time= 0.5s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=325; total time= 0.4s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=325; total time= 0.5s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=325; total time= 0.4s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=325; total time= 0.4s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=325; total time= 0.4s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=292; total time= 0.4s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=292; total time= 0.4s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=292; total time= 0.3s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=292; total time= 0.4s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=292; total time= 0.3s [CV] END bootstrap=False, max_depth=110, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=988; total time= 1.1s [CV] END bootstrap=False, max_depth=110, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=988; total time= 1.0s [CV] END bootstrap=False, max_depth=110, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=988; total time= 1.0s [CV] END bootstrap=False, max_depth=110, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=988; total time= 1.2s [CV] END bootstrap=False, max_depth=110, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=988; total time= 1.0s [CV] END bootstrap=False, max_depth=210, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=631; total time= 0.8s [CV] END bootstrap=False, max_depth=210, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=631; total time= 0.7s [CV] END bootstrap=False, max_depth=210, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=631; total time= 0.7s [CV] END bootstrap=False, max_depth=210, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=631; total time= 0.8s [CV] END bootstrap=False, max_depth=210, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=631; total time= 0.7s [CV] END bootstrap=False, max_depth=185, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=700; total time= 0.9s [CV] END bootstrap=False, max_depth=185, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=700; total time= 0.7s [CV] END bootstrap=False, max_depth=185, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=700; total time= 0.8s [CV] END bootstrap=False, max_depth=185, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=700; total time= 0.8s [CV] END bootstrap=False, max_depth=185, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=700; total time= 0.8s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=610; total time= 0.7s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=610; total time= 0.8s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=610; total time= 0.7s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=610; total time= 0.6s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=610; total time= 0.7s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=877; total time= 1.0s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=877; total time= 1.0s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=877; total time= 0.9s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=877; total time= 1.0s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=877; total time= 1.0s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=580; total time= 0.6s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=580; total time= 0.6s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=580; total time= 0.6s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=580; total time= 0.6s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=580; total time= 0.6s [CV] END bootstrap=False, max_depth=115, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=25; total time= 0.0s [CV] END bootstrap=False, max_depth=115, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=25; total time= 0.0s [CV] END bootstrap=False, max_depth=115, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=25; total time= 0.0s [CV] END bootstrap=False, max_depth=115, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=25; total time= 0.0s [CV] END bootstrap=False, max_depth=115, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=25; total time= 0.0s [CV] END bootstrap=True, max_depth=50, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=286; total time= 0.3s [CV] END bootstrap=True, max_depth=50, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=286; total time= 0.3s [CV] END bootstrap=True, max_depth=50, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=286; total time= 0.3s [CV] END bootstrap=True, max_depth=50, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=286; total time= 0.3s [CV] END bootstrap=True, max_depth=50, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=286; total time= 0.3s [CV] END bootstrap=False, max_depth=40, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=742; total time= 0.8s [CV] END bootstrap=False, max_depth=40, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=742; total time= 0.8s [CV] END bootstrap=False, max_depth=40, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=742; total time= 1.0s [CV] END bootstrap=False, max_depth=40, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=742; total time= 0.8s [CV] END bootstrap=False, max_depth=40, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=742; total time= 0.7s [CV] END bootstrap=False, max_depth=155, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=793; total time= 0.7s [CV] END bootstrap=False, max_depth=155, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=793; total time= 0.8s [CV] END bootstrap=False, max_depth=155, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=793; total time= 0.7s [CV] END bootstrap=False, max_depth=155, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=793; total time= 0.8s [CV] END bootstrap=False, max_depth=155, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=793; total time= 0.8s [CV] END bootstrap=False, max_depth=70, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=61; total time= 0.0s [CV] END bootstrap=False, max_depth=70, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=61; total time= 0.0s [CV] END bootstrap=False, max_depth=70, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=61; total time= 0.0s [CV] END bootstrap=False, max_depth=70, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=61; total time= 0.0s [CV] END bootstrap=False, max_depth=70, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=61; total time= 0.1s [CV] END bootstrap=True, max_depth=175, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=457; total time= 0.6s [CV] END bootstrap=True, max_depth=175, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=457; total time= 0.5s [CV] END bootstrap=True, max_depth=175, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=457; total time= 0.5s [CV] END bootstrap=True, max_depth=175, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=457; total time= 0.5s [CV] END bootstrap=True, max_depth=175, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=457; total time= 0.6s [CV] END bootstrap=True, max_depth=20, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=334; total time= 0.4s [CV] END bootstrap=True, max_depth=20, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=334; total time= 0.4s [CV] END bootstrap=True, max_depth=20, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=334; total time= 0.5s [CV] END bootstrap=True, max_depth=20, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=334; total time= 0.5s [CV] END bootstrap=True, max_depth=20, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=334; total time= 0.4s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=379; total time= 0.3s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=379; total time= 0.3s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=379; total time= 0.4s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=379; total time= 0.2s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=379; total time= 0.3s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=718; total time= 0.8s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=718; total time= 0.8s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=718; total time= 0.8s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=718; total time= 0.8s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=718; total time= 0.8s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=952; total time= 1.0s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=952; total time= 0.9s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=952; total time= 1.0s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=952; total time= 1.1s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=952; total time= 0.9s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=739; total time= 0.8s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=739; total time= 0.8s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=739; total time= 0.7s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=739; total time= 0.8s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=739; total time= 0.8s [CV] END bootstrap=True, max_depth=50, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=745; total time= 0.8s [CV] END bootstrap=True, max_depth=50, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=745; total time= 0.8s [CV] END bootstrap=True, max_depth=50, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=745; total time= 0.8s [CV] END bootstrap=True, max_depth=50, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=745; total time= 0.7s [CV] END bootstrap=True, max_depth=50, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=745; total time= 0.7s [CV] END bootstrap=False, max_depth=225, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=940; total time= 1.0s [CV] END bootstrap=False, max_depth=225, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=940; total time= 1.0s [CV] END bootstrap=False, max_depth=225, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=940; total time= 1.0s [CV] END bootstrap=False, max_depth=225, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=940; total time= 1.0s [CV] END bootstrap=False, max_depth=225, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=940; total time= 1.0s [CV] END bootstrap=False, max_depth=200, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=25; total time= 0.0s [CV] END bootstrap=False, max_depth=200, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=25; total time= 0.0s [CV] END bootstrap=False, max_depth=200, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=25; total time= 0.0s [CV] END bootstrap=False, max_depth=200, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=25; total time= 0.0s [CV] END bootstrap=False, max_depth=200, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=25; total time= 0.0s [CV] END bootstrap=True, max_depth=240, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=256; total time= 0.3s [CV] END bootstrap=True, max_depth=240, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=256; total time= 0.2s [CV] END bootstrap=True, max_depth=240, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=256; total time= 0.2s [CV] END bootstrap=True, max_depth=240, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=256; total time= 0.2s [CV] END bootstrap=True, max_depth=240, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=256; total time= 0.2s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=967; total time= 1.0s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=967; total time= 0.9s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=967; total time= 1.1s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=967; total time= 1.0s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=967; total time= 0.9s [CV] END bootstrap=False, max_depth=185, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=718; total time= 0.8s [CV] END bootstrap=False, max_depth=185, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=718; total time= 0.8s [CV] END bootstrap=False, max_depth=185, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=718; total time= 0.9s [CV] END bootstrap=False, max_depth=185, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=718; total time= 0.8s [CV] END bootstrap=False, max_depth=185, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=718; total time= 0.7s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=604; total time= 0.7s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=604; total time= 0.8s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=604; total time= 0.7s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=604; total time= 0.8s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=604; total time= 0.7s [CV] END bootstrap=True, max_depth=220, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=616; total time= 0.9s [CV] END bootstrap=True, max_depth=220, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=616; total time= 0.7s [CV] END bootstrap=True, max_depth=220, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=616; total time= 0.8s [CV] END bootstrap=True, max_depth=220, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=616; total time= 0.7s [CV] END bootstrap=True, max_depth=220, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=616; total time= 0.7s [CV] END bootstrap=False, max_depth=60, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=844; total time= 0.9s [CV] END bootstrap=False, max_depth=60, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=844; total time= 0.8s [CV] END bootstrap=False, max_depth=60, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=844; total time= 1.0s [CV] END bootstrap=False, max_depth=60, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=844; total time= 0.9s [CV] END bootstrap=False, max_depth=60, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=844; total time= 0.9s [CV] END bootstrap=True, max_depth=210, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=868; total time= 0.9s [CV] END bootstrap=True, max_depth=210, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=868; total time= 1.0s [CV] END bootstrap=True, max_depth=210, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=868; total time= 1.0s [CV] END bootstrap=True, max_depth=210, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=868; total time= 1.0s [CV] END bootstrap=True, max_depth=210, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=868; total time= 0.9s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=307; total time= 0.4s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=307; total time= 0.4s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=307; total time= 0.4s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=307; total time= 0.4s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=307; total time= 0.3s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=829; total time= 0.8s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=829; total time= 0.8s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=829; total time= 0.9s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=829; total time= 1.0s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=829; total time= 0.8s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=607; total time= 0.7s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=607; total time= 0.7s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=607; total time= 0.7s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=607; total time= 0.7s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=607; total time= 0.8s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=688; total time= 0.8s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=688; total time= 0.8s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=688; total time= 0.7s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=688; total time= 0.7s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=688; total time= 0.8s [CV] END bootstrap=True, max_depth=35, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=13; total time= 0.0s [CV] END bootstrap=True, max_depth=35, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=13; total time= 0.0s [CV] END bootstrap=True, max_depth=35, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=13; total time= 0.0s [CV] END bootstrap=True, max_depth=35, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=13; total time= 0.0s [CV] END bootstrap=True, max_depth=35, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=13; total time= 0.0s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=97; total time= 0.1s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=97; total time= 0.1s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=97; total time= 0.1s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=97; total time= 0.1s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=97; total time= 0.1s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=595; total time= 0.7s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=595; total time= 0.6s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=595; total time= 0.6s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=595; total time= 0.6s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=595; total time= 0.6s [CV] END bootstrap=False, max_depth=135, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=751; total time= 0.7s [CV] END bootstrap=False, max_depth=135, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=751; total time= 0.8s [CV] END bootstrap=False, max_depth=135, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=751; total time= 0.8s [CV] END bootstrap=False, max_depth=135, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=751; total time= 0.8s [CV] END bootstrap=False, max_depth=135, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=751; total time= 0.8s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=952; total time= 1.0s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=952; total time= 0.9s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=952; total time= 1.0s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=952; total time= 1.0s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=952; total time= 1.0s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=523; total time= 0.5s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=523; total time= 0.4s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=523; total time= 0.5s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=523; total time= 0.4s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=523; total time= 0.5s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=496; total time= 0.4s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=496; total time= 0.4s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=496; total time= 0.4s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=496; total time= 0.3s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=496; total time= 0.5s [CV] END bootstrap=False, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=697; total time= 0.8s [CV] END bootstrap=False, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=697; total time= 0.8s [CV] END bootstrap=False, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=697; total time= 0.7s [CV] END bootstrap=False, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=697; total time= 0.7s [CV] END bootstrap=False, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=697; total time= 0.7s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=19; total time= 0.0s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=19; total time= 0.0s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=19; total time= 0.0s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=19; total time= 0.0s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=19; total time= 0.0s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=172; total time= 0.3s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=172; total time= 0.3s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=172; total time= 0.2s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=172; total time= 0.3s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=172; total time= 0.2s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=103; total time= 0.1s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=103; total time= 0.2s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=103; total time= 0.1s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=103; total time= 0.1s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=103; total time= 0.1s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=388; total time= 0.4s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=388; total time= 0.3s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=388; total time= 0.3s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=388; total time= 0.4s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=388; total time= 0.4s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=139; total time= 0.2s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=139; total time= 0.2s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=139; total time= 0.2s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=139; total time= 0.1s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=139; total time= 0.2s [CV] END bootstrap=False, max_depth=225, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=430; total time= 0.5s [CV] END bootstrap=False, max_depth=225, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=430; total time= 0.5s [CV] END bootstrap=False, max_depth=225, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=430; total time= 0.5s [CV] END bootstrap=False, max_depth=225, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=430; total time= 0.5s [CV] END bootstrap=False, max_depth=225, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=430; total time= 0.5s [CV] END bootstrap=False, max_depth=40, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=28; total time= 0.0s [CV] END bootstrap=False, max_depth=40, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=28; total time= 0.0s [CV] END bootstrap=False, max_depth=40, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=28; total time= 0.0s [CV] END bootstrap=False, max_depth=40, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=28; total time= 0.0s [CV] END bootstrap=False, max_depth=40, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=28; total time= 0.0s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=445; total time= 0.5s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=445; total time= 0.5s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=445; total time= 0.6s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=445; total time= 0.5s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=445; total time= 0.4s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=760; total time= 0.6s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=760; total time= 0.6s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=760; total time= 0.6s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=760; total time= 0.7s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=760; total time= 0.6s [CV] END bootstrap=True, max_depth=230, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=643; total time= 0.8s [CV] END bootstrap=True, max_depth=230, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=643; total time= 0.8s [CV] END bootstrap=True, max_depth=230, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=643; total time= 0.7s [CV] END bootstrap=True, max_depth=230, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=643; total time= 0.8s [CV] END bootstrap=True, max_depth=230, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=643; total time= 0.8s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=979; total time= 0.9s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=979; total time= 1.1s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=979; total time= 1.1s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=979; total time= 1.1s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=979; total time= 1.0s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=247; total time= 0.3s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=247; total time= 0.3s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=247; total time= 0.2s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=247; total time= 0.2s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=247; total time= 0.3s [CV] END bootstrap=True, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=658; total time= 0.7s [CV] END bootstrap=True, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=658; total time= 0.8s [CV] END bootstrap=True, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=658; total time= 0.6s [CV] END bootstrap=True, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=658; total time= 0.7s [CV] END bootstrap=True, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=658; total time= 0.8s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=502; total time= 0.4s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=502; total time= 0.5s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=502; total time= 0.3s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=502; total time= 0.5s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=502; total time= 0.6s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=706; total time= 0.9s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=706; total time= 0.9s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=706; total time= 0.8s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=706; total time= 0.9s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=706; total time= 0.8s [CV] END bootstrap=True, max_depth=45, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=67; total time= 0.1s [CV] END bootstrap=True, max_depth=45, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=67; total time= 0.1s [CV] END bootstrap=True, max_depth=45, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=67; total time= 0.1s [CV] END bootstrap=True, max_depth=45, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=67; total time= 0.0s [CV] END bootstrap=True, max_depth=45, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=67; total time= 0.1s [CV] END bootstrap=True, max_depth=130, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=898; total time= 1.0s [CV] END bootstrap=True, max_depth=130, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=898; total time= 1.1s [CV] END bootstrap=True, max_depth=130, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=898; total time= 1.1s [CV] END bootstrap=True, max_depth=130, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=898; total time= 1.0s [CV] END bootstrap=True, max_depth=130, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=898; total time= 1.0s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=559; total time= 0.6s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=559; total time= 0.5s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=559; total time= 0.5s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=559; total time= 0.5s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=559; total time= 0.6s [CV] END bootstrap=False, max_depth=80, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=52; total time= 0.1s [CV] END bootstrap=False, max_depth=80, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=52; total time= 0.0s [CV] END bootstrap=False, max_depth=80, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=52; total time= 0.0s [CV] END bootstrap=False, max_depth=80, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=52; total time= 0.0s [CV] END bootstrap=False, max_depth=80, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=52; total time= 0.0s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=97; total time= 0.1s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=97; total time= 0.1s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=97; total time= 0.1s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=97; total time= 0.1s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=97; total time= 0.1s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=964; total time= 0.8s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=964; total time= 1.0s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=964; total time= 0.9s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=964; total time= 1.0s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=964; total time= 1.0s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=169; total time= 0.3s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=169; total time= 0.2s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=169; total time= 0.2s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=169; total time= 0.3s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=169; total time= 0.3s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=352; total time= 0.5s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=352; total time= 0.4s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=352; total time= 0.5s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=352; total time= 0.5s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=352; total time= 0.5s [CV] END bootstrap=True, max_depth=15, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=925; total time= 1.0s [CV] END bootstrap=True, max_depth=15, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=925; total time= 0.9s [CV] END bootstrap=True, max_depth=15, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=925; total time= 1.0s [CV] END bootstrap=True, max_depth=15, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=925; total time= 0.9s [CV] END bootstrap=True, max_depth=15, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=925; total time= 0.9s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=364; total time= 0.6s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=364; total time= 0.6s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=364; total time= 0.6s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=364; total time= 0.5s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=364; total time= 0.4s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=826; total time= 0.8s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=826; total time= 0.8s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=826; total time= 0.8s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=826; total time= 0.8s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=826; total time= 0.8s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=691; total time= 0.7s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=691; total time= 0.7s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=691; total time= 0.9s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=691; total time= 0.8s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=691; total time= 0.8s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=745; total time= 0.9s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=745; total time= 0.7s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=745; total time= 0.8s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=745; total time= 0.7s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=745; total time= 0.7s [CV] END bootstrap=True, max_depth=45, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=526; total time= 0.5s [CV] END bootstrap=True, max_depth=45, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=526; total time= 0.5s [CV] END bootstrap=True, max_depth=45, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=526; total time= 0.4s [CV] END bootstrap=True, max_depth=45, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=526; total time= 0.5s [CV] END bootstrap=True, max_depth=45, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=526; total time= 0.4s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=745; total time= 0.7s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=745; total time= 0.7s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=745; total time= 0.7s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=745; total time= 0.7s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=745; total time= 0.8s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=490; total time= 0.6s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=490; total time= 0.6s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=490; total time= 0.7s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=490; total time= 0.6s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=490; total time= 0.6s [CV] END bootstrap=False, max_depth=65, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=106; total time= 0.1s [CV] END bootstrap=False, max_depth=65, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=106; total time= 0.1s [CV] END bootstrap=False, max_depth=65, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=106; total time= 0.1s [CV] END bootstrap=False, max_depth=65, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=106; total time= 0.2s [CV] END bootstrap=False, max_depth=65, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=106; total time= 0.2s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=871; total time= 0.9s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=871; total time= 1.0s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=871; total time= 0.9s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=871; total time= 0.9s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=871; total time= 0.9s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=871; total time= 0.9s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=871; total time= 1.1s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=871; total time= 1.0s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=871; total time= 1.1s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=871; total time= 0.9s [CV] END bootstrap=True, max_depth=235, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=979; total time= 1.0s [CV] END bootstrap=True, max_depth=235, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=979; total time= 1.1s [CV] END bootstrap=True, max_depth=235, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=979; total time= 1.0s [CV] END bootstrap=True, max_depth=235, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=979; total time= 1.0s [CV] END bootstrap=True, max_depth=235, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=979; total time= 1.0s [CV] END bootstrap=False, max_depth=65, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=1000; total time= 1.0s [CV] END bootstrap=False, max_depth=65, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=1000; total time= 0.9s [CV] END bootstrap=False, max_depth=65, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=1000; total time= 0.9s [CV] END bootstrap=False, max_depth=65, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=1000; total time= 0.9s [CV] END bootstrap=False, max_depth=65, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=1000; total time= 1.0s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=550; total time= 0.5s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=550; total time= 0.5s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=550; total time= 0.5s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=550; total time= 0.6s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=550; total time= 0.5s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=109; total time= 0.1s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=109; total time= 0.2s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=109; total time= 0.2s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=109; total time= 0.1s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=109; total time= 0.1s [CV] END bootstrap=False, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=967; total time= 1.0s [CV] END bootstrap=False, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=967; total time= 1.1s [CV] END bootstrap=False, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=967; total time= 1.1s [CV] END bootstrap=False, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=967; total time= 1.0s [CV] END bootstrap=False, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=967; total time= 1.1s [CV] END bootstrap=False, max_depth=25, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=457; total time= 0.6s [CV] END bootstrap=False, max_depth=25, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=457; total time= 0.6s [CV] END bootstrap=False, max_depth=25, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=457; total time= 0.6s [CV] END bootstrap=False, max_depth=25, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=457; total time= 0.6s [CV] END bootstrap=False, max_depth=25, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=457; total time= 0.5s [CV] END bootstrap=True, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=154; total time= 0.2s [CV] END bootstrap=True, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=154; total time= 0.2s [CV] END bootstrap=True, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=154; total time= 0.2s [CV] END bootstrap=True, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=154; total time= 0.3s [CV] END bootstrap=True, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=154; total time= 0.2s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=946; total time= 1.0s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=946; total time= 0.9s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=946; total time= 0.9s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=946; total time= 1.0s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=946; total time= 0.9s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=973; total time= 1.1s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=973; total time= 1.1s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=973; total time= 1.1s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=973; total time= 1.0s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=973; total time= 0.9s [CV] END bootstrap=True, max_depth=140, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=772; total time= 0.6s [CV] END bootstrap=True, max_depth=140, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=772; total time= 0.7s [CV] END bootstrap=True, max_depth=140, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=772; total time= 0.7s [CV] END bootstrap=True, max_depth=140, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=772; total time= 0.7s [CV] END bootstrap=True, max_depth=140, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=772; total time= 0.7s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=676; total time= 0.7s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=676; total time= 0.7s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=676; total time= 0.7s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=676; total time= 0.7s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=676; total time= 0.7s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=250; total time= 0.2s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=250; total time= 0.2s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=250; total time= 0.3s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=250; total time= 0.2s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=250; total time= 0.2s [CV] END bootstrap=True, max_depth=230, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=994; total time= 0.9s [CV] END bootstrap=True, max_depth=230, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=994; total time= 0.9s [CV] END bootstrap=True, max_depth=230, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=994; total time= 1.1s [CV] END bootstrap=True, max_depth=230, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=994; total time= 1.0s [CV] END bootstrap=True, max_depth=230, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=994; total time= 1.0s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=163; total time= 0.2s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=163; total time= 0.2s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=163; total time= 0.3s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=163; total time= 0.2s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=163; total time= 0.3s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=586; total time= 0.6s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=586; total time= 0.7s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=586; total time= 0.6s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=586; total time= 0.8s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=586; total time= 0.7s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=649; total time= 0.8s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=649; total time= 0.8s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=649; total time= 0.8s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=649; total time= 0.9s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=649; total time= 0.8s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=199; total time= 0.2s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=199; total time= 0.2s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=199; total time= 0.2s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=199; total time= 0.2s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=199; total time= 0.3s [CV] END bootstrap=False, max_depth=155, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=274; total time= 0.3s [CV] END bootstrap=False, max_depth=155, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=274; total time= 0.3s [CV] END bootstrap=False, max_depth=155, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=274; total time= 0.3s [CV] END bootstrap=False, max_depth=155, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=274; total time= 0.3s [CV] END bootstrap=False, max_depth=155, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=274; total time= 0.2s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=850; total time= 0.9s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=850; total time= 1.0s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=850; total time= 0.9s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=850; total time= 0.8s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=850; total time= 0.8s [CV] END bootstrap=False, max_depth=35, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=190; total time= 0.2s [CV] END bootstrap=False, max_depth=35, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=190; total time= 0.2s [CV] END bootstrap=False, max_depth=35, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=190; total time= 0.2s [CV] END bootstrap=False, max_depth=35, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=190; total time= 0.2s [CV] END bootstrap=False, max_depth=35, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=190; total time= 0.2s [CV] END bootstrap=False, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=877; total time= 1.0s [CV] END bootstrap=False, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=877; total time= 1.0s [CV] END bootstrap=False, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=877; total time= 1.0s [CV] END bootstrap=False, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=877; total time= 1.1s [CV] END bootstrap=False, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=877; total time= 1.0s [CV] END bootstrap=False, max_depth=15, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=709; total time= 0.8s [CV] END bootstrap=False, max_depth=15, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=709; total time= 0.8s [CV] END bootstrap=False, max_depth=15, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=709; total time= 0.8s [CV] END bootstrap=False, max_depth=15, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=709; total time= 0.8s [CV] END bootstrap=False, max_depth=15, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=709; total time= 0.8s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=634; total time= 0.8s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=634; total time= 0.7s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=634; total time= 0.8s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=634; total time= 0.6s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=634; total time= 0.8s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=334; total time= 0.5s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=334; total time= 0.5s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=334; total time= 0.4s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=334; total time= 0.5s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=334; total time= 0.5s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=25; total time= 0.0s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=25; total time= 0.0s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=25; total time= 0.0s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=25; total time= 0.0s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=25; total time= 0.0s [CV] END bootstrap=True, max_depth=45, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=595; total time= 0.7s [CV] END bootstrap=True, max_depth=45, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=595; total time= 0.7s [CV] END bootstrap=True, max_depth=45, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=595; total time= 0.6s [CV] END bootstrap=True, max_depth=45, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=595; total time= 0.6s [CV] END bootstrap=True, max_depth=45, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=595; total time= 0.7s [CV] END bootstrap=True, max_depth=135, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=25; total time= 0.0s [CV] END bootstrap=True, max_depth=135, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=25; total time= 0.0s [CV] END bootstrap=True, max_depth=135, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=25; total time= 0.0s [CV] END bootstrap=True, max_depth=135, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=25; total time= 0.0s [CV] END bootstrap=True, max_depth=135, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=25; total time= 0.0s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=514; total time= 0.4s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=514; total time= 0.5s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=514; total time= 0.4s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=514; total time= 0.4s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=514; total time= 0.4s [CV] END bootstrap=True, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=820; total time= 0.8s [CV] END bootstrap=True, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=820; total time= 0.8s [CV] END bootstrap=True, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=820; total time= 0.9s [CV] END bootstrap=True, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=820; total time= 0.7s [CV] END bootstrap=True, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=820; total time= 0.7s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=769; total time= 0.6s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=769; total time= 0.6s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=769; total time= 0.6s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=769; total time= 0.7s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=769; total time= 0.6s [CV] END bootstrap=False, max_depth=70, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=592; total time= 0.7s [CV] END bootstrap=False, max_depth=70, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=592; total time= 0.7s [CV] END bootstrap=False, max_depth=70, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=592; total time= 0.6s [CV] END bootstrap=False, max_depth=70, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=592; total time= 0.6s [CV] END bootstrap=False, max_depth=70, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=592; total time= 0.6s [CV] END bootstrap=True, max_depth=30, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=889; total time= 1.1s [CV] END bootstrap=True, max_depth=30, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=889; total time= 0.9s [CV] END bootstrap=True, max_depth=30, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=889; total time= 1.0s [CV] END bootstrap=True, max_depth=30, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=889; total time= 1.0s [CV] END bootstrap=True, max_depth=30, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=889; total time= 1.0s [CV] END bootstrap=False, max_depth=30, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=526; total time= 0.5s [CV] END bootstrap=False, max_depth=30, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=526; total time= 0.5s [CV] END bootstrap=False, max_depth=30, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=526; total time= 0.4s [CV] END bootstrap=False, max_depth=30, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=526; total time= 0.5s [CV] END bootstrap=False, max_depth=30, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=526; total time= 0.5s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=502; total time= 0.4s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=502; total time= 0.4s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=502; total time= 0.6s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=502; total time= 0.5s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=502; total time= 0.8s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=316; total time= 0.3s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=316; total time= 0.4s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=316; total time= 0.4s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=316; total time= 0.5s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=316; total time= 0.3s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=649; total time= 0.9s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=649; total time= 0.9s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=649; total time= 0.9s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=649; total time= 0.8s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=649; total time= 0.8s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=985; total time= 1.0s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=985; total time= 1.1s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=985; total time= 1.0s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=985; total time= 1.0s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=985; total time= 1.0s [CV] END bootstrap=False, max_depth=190, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=337; total time= 0.6s [CV] END bootstrap=False, max_depth=190, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=337; total time= 0.5s [CV] END bootstrap=False, max_depth=190, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=337; total time= 0.5s [CV] END bootstrap=False, max_depth=190, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=337; total time= 0.5s [CV] END bootstrap=False, max_depth=190, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=337; total time= 0.5s [CV] END bootstrap=False, max_depth=25, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=232; total time= 0.3s [CV] END bootstrap=False, max_depth=25, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=232; total time= 0.3s [CV] END bootstrap=False, max_depth=25, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=232; total time= 0.3s [CV] END bootstrap=False, max_depth=25, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=232; total time= 0.3s [CV] END bootstrap=False, max_depth=25, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=232; total time= 0.3s [CV] END bootstrap=False, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=499; total time= 0.6s [CV] END bootstrap=False, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=499; total time= 0.4s [CV] END bootstrap=False, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=499; total time= 0.6s [CV] END bootstrap=False, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=499; total time= 0.3s [CV] END bootstrap=False, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=499; total time= 0.5s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=904; total time= 1.0s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=904; total time= 1.0s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=904; total time= 1.0s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=904; total time= 1.0s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=904; total time= 1.1s [CV] END bootstrap=False, max_depth=140, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=172; total time= 0.3s [CV] END bootstrap=False, max_depth=140, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=172; total time= 0.3s [CV] END bootstrap=False, max_depth=140, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=172; total time= 0.3s [CV] END bootstrap=False, max_depth=140, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=172; total time= 0.2s [CV] END bootstrap=False, max_depth=140, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=172; total time= 0.2s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=325; total time= 0.5s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=325; total time= 0.4s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=325; total time= 0.5s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=325; total time= 0.5s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=325; total time= 0.5s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=571; total time= 0.5s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=571; total time= 0.6s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=571; total time= 0.5s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=571; total time= 0.5s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=571; total time= 0.6s [CV] END bootstrap=False, max_depth=25, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=49; total time= 0.0s [CV] END bootstrap=False, max_depth=25, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=49; total time= 0.0s [CV] END bootstrap=False, max_depth=25, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=49; total time= 0.0s [CV] END bootstrap=False, max_depth=25, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=49; total time= 0.0s [CV] END bootstrap=False, max_depth=25, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=49; total time= 0.0s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=133; total time= 0.2s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=133; total time= 0.2s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=133; total time= 0.2s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=133; total time= 0.1s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=133; total time= 0.1s [CV] END bootstrap=True, max_depth=135, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=112; total time= 0.2s [CV] END bootstrap=True, max_depth=135, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=112; total time= 0.1s [CV] END bootstrap=True, max_depth=135, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=112; total time= 0.1s [CV] END bootstrap=True, max_depth=135, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=112; total time= 0.1s [CV] END bootstrap=True, max_depth=135, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=112; total time= 0.1s [CV] END bootstrap=False, max_depth=230, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=268; total time= 0.3s [CV] END bootstrap=False, max_depth=230, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=268; total time= 0.2s [CV] END bootstrap=False, max_depth=230, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=268; total time= 0.2s [CV] END bootstrap=False, max_depth=230, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=268; total time= 0.4s [CV] END bootstrap=False, max_depth=230, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=268; total time= 0.3s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=532; total time= 0.6s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=532; total time= 0.5s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=532; total time= 0.5s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=532; total time= 0.5s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=532; total time= 0.4s [CV] END bootstrap=True, max_depth=20, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=172; total time= 0.3s [CV] END bootstrap=True, max_depth=20, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=172; total time= 0.2s [CV] END bootstrap=True, max_depth=20, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=172; total time= 0.3s [CV] END bootstrap=True, max_depth=20, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=172; total time= 0.3s [CV] END bootstrap=True, max_depth=20, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=172; total time= 0.2s [CV] END bootstrap=True, max_depth=80, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=127; total time= 0.2s [CV] END bootstrap=True, max_depth=80, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=127; total time= 0.1s [CV] END bootstrap=True, max_depth=80, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=127; total time= 0.2s [CV] END bootstrap=True, max_depth=80, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=127; total time= 0.1s [CV] END bootstrap=True, max_depth=80, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=127; total time= 0.1s [CV] END bootstrap=True, max_depth=130, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=970; total time= 1.0s [CV] END bootstrap=True, max_depth=130, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=970; total time= 1.0s [CV] END bootstrap=True, max_depth=130, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=970; total time= 1.1s [CV] END bootstrap=True, max_depth=130, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=970; total time= 1.0s [CV] END bootstrap=True, max_depth=130, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=970; total time= 0.9s [CV] END bootstrap=True, max_depth=75, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=859; total time= 0.9s [CV] END bootstrap=True, max_depth=75, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=859; total time= 1.0s [CV] END bootstrap=True, max_depth=75, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=859; total time= 0.9s [CV] END bootstrap=True, max_depth=75, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=859; total time= 0.9s [CV] END bootstrap=True, max_depth=75, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=859; total time= 0.9s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=868; total time= 0.9s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=868; total time= 0.9s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=868; total time= 0.9s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=868; total time= 0.9s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=868; total time= 0.9s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=745; total time= 0.8s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=745; total time= 0.7s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=745; total time= 0.8s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=745; total time= 0.7s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=745; total time= 0.8s [CV] END bootstrap=False, max_depth=55, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=556; total time= 0.5s [CV] END bootstrap=False, max_depth=55, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=556; total time= 0.6s [CV] END bootstrap=False, max_depth=55, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=556; total time= 0.6s [CV] END bootstrap=False, max_depth=55, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=556; total time= 0.6s [CV] END bootstrap=False, max_depth=55, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=556; total time= 0.5s [CV] END bootstrap=True, max_depth=160, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=943; total time= 1.0s [CV] END bootstrap=True, max_depth=160, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=943; total time= 0.9s [CV] END bootstrap=True, max_depth=160, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=943; total time= 1.0s [CV] END bootstrap=True, max_depth=160, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=943; total time= 1.0s [CV] END bootstrap=True, max_depth=160, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=943; total time= 0.9s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=34; total time= 0.0s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=34; total time= 0.0s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=34; total time= 0.0s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=34; total time= 0.0s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=34; total time= 0.0s [CV] END bootstrap=True, max_depth=15, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=994; total time= 0.9s [CV] END bootstrap=True, max_depth=15, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=994; total time= 0.9s [CV] END bootstrap=True, max_depth=15, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=994; total time= 0.9s [CV] END bootstrap=True, max_depth=15, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=994; total time= 1.0s [CV] END bootstrap=True, max_depth=15, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=994; total time= 1.0s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=532; total time= 0.4s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=532; total time= 0.5s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=532; total time= 0.4s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=532; total time= 0.5s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=532; total time= 0.7s [CV] END bootstrap=False, max_depth=135, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=970; total time= 1.0s [CV] END bootstrap=False, max_depth=135, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=970; total time= 1.1s [CV] END bootstrap=False, max_depth=135, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=970; total time= 1.1s [CV] END bootstrap=False, max_depth=135, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=970; total time= 1.1s [CV] END bootstrap=False, max_depth=135, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=970; total time= 1.0s [CV] END bootstrap=False, max_depth=70, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=673; total time= 0.8s [CV] END bootstrap=False, max_depth=70, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=673; total time= 0.7s [CV] END bootstrap=False, max_depth=70, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=673; total time= 0.8s [CV] END bootstrap=False, max_depth=70, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=673; total time= 0.8s [CV] END bootstrap=False, max_depth=70, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=673; total time= 0.7s [CV] END bootstrap=True, max_depth=30, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=76; total time= 0.1s [CV] END bootstrap=True, max_depth=30, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=76; total time= 0.1s [CV] END bootstrap=True, max_depth=30, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=76; total time= 0.1s [CV] END bootstrap=True, max_depth=30, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=76; total time= 0.1s [CV] END bootstrap=True, max_depth=30, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=76; total time= 0.1s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=229; total time= 0.3s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=229; total time= 0.2s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=229; total time= 0.2s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=229; total time= 0.2s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=229; total time= 0.3s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=181; total time= 0.2s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=181; total time= 0.2s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=181; total time= 0.2s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=181; total time= 0.1s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=181; total time= 0.3s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=529; total time= 0.5s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=529; total time= 0.5s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=529; total time= 0.5s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=529; total time= 0.4s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=529; total time= 0.5s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=916; total time= 1.2s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=916; total time= 1.0s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=916; total time= 0.8s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=916; total time= 1.0s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=916; total time= 0.8s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=829; total time= 0.7s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=829; total time= 0.9s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=829; total time= 0.8s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=829; total time= 0.8s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=829; total time= 0.9s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=28; total time= 0.0s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=28; total time= 0.0s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=28; total time= 0.0s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=28; total time= 0.0s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=28; total time= 0.0s [CV] END bootstrap=False, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=460; total time= 0.6s [CV] END bootstrap=False, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=460; total time= 0.6s [CV] END bootstrap=False, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=460; total time= 0.6s [CV] END bootstrap=False, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=460; total time= 0.5s [CV] END bootstrap=False, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=460; total time= 0.5s [CV] END bootstrap=False, max_depth=215, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=493; total time= 0.7s [CV] END bootstrap=False, max_depth=215, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=493; total time= 0.7s [CV] END bootstrap=False, max_depth=215, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=493; total time= 0.7s [CV] END bootstrap=False, max_depth=215, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=493; total time= 0.6s [CV] END bootstrap=False, max_depth=215, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=493; total time= 0.7s [CV] END bootstrap=True, max_depth=30, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=748; total time= 0.9s [CV] END bootstrap=True, max_depth=30, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=748; total time= 0.7s [CV] END bootstrap=True, max_depth=30, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=748; total time= 0.9s [CV] END bootstrap=True, max_depth=30, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=748; total time= 0.6s [CV] END bootstrap=True, max_depth=30, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=748; total time= 0.8s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=550; total time= 0.7s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=550; total time= 0.6s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=550; total time= 0.6s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=550; total time= 0.6s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=550; total time= 0.5s [CV] END bootstrap=True, max_depth=230, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=421; total time= 0.4s [CV] END bootstrap=True, max_depth=230, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=421; total time= 0.5s [CV] END bootstrap=True, max_depth=230, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=421; total time= 0.4s [CV] END bootstrap=True, max_depth=230, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=421; total time= 0.4s [CV] END bootstrap=True, max_depth=230, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=421; total time= 0.5s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=541; total time= 0.5s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=541; total time= 0.5s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=541; total time= 0.5s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=541; total time= 0.6s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=541; total time= 0.4s [CV] END bootstrap=True, max_depth=15, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=487; total time= 0.7s [CV] END bootstrap=True, max_depth=15, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=487; total time= 0.7s [CV] END bootstrap=True, max_depth=15, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=487; total time= 0.7s [CV] END bootstrap=True, max_depth=15, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=487; total time= 0.7s [CV] END bootstrap=True, max_depth=15, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=487; total time= 0.6s [CV] END bootstrap=False, max_depth=170, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=829; total time= 0.8s [CV] END bootstrap=False, max_depth=170, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=829; total time= 0.8s [CV] END bootstrap=False, max_depth=170, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=829; total time= 0.9s [CV] END bootstrap=False, max_depth=170, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=829; total time= 0.8s [CV] END bootstrap=False, max_depth=170, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=829; total time= 0.9s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=706; total time= 0.7s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=706; total time= 0.9s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=706; total time= 0.8s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=706; total time= 0.8s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=706; total time= 0.8s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=361; total time= 0.5s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=361; total time= 0.6s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=361; total time= 0.4s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=361; total time= 0.6s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=361; total time= 0.5s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=100; total time= 0.1s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=100; total time= 0.1s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=100; total time= 0.1s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=100; total time= 0.1s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=100; total time= 0.2s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=736; total time= 0.8s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=736; total time= 0.7s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=736; total time= 0.9s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=736; total time= 0.7s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=736; total time= 0.8s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=556; total time= 0.8s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=556; total time= 0.6s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=556; total time= 0.5s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=556; total time= 0.6s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=556; total time= 0.6s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=727; total time= 0.8s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=727; total time= 0.9s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=727; total time= 0.8s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=727; total time= 0.8s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=727; total time= 0.9s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=820; total time= 0.6s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=820; total time= 0.7s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=820; total time= 0.7s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=820; total time= 0.7s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=820; total time= 0.7s [CV] END bootstrap=False, max_depth=115, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=895; total time= 0.9s [CV] END bootstrap=False, max_depth=115, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=895; total time= 0.9s [CV] END bootstrap=False, max_depth=115, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=895; total time= 0.9s [CV] END bootstrap=False, max_depth=115, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=895; total time= 0.9s [CV] END bootstrap=False, max_depth=115, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=895; total time= 1.0s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=22; total time= 0.0s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=22; total time= 0.0s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=22; total time= 0.0s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=22; total time= 0.0s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=22; total time= 0.0s [CV] END bootstrap=False, max_depth=60, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=466; total time= 0.7s [CV] END bootstrap=False, max_depth=60, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=466; total time= 0.6s [CV] END bootstrap=False, max_depth=60, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=466; total time= 0.8s [CV] END bootstrap=False, max_depth=60, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=466; total time= 0.8s [CV] END bootstrap=False, max_depth=60, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=466; total time= 0.8s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=628; total time= 0.9s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=628; total time= 1.1s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=628; total time= 1.1s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=628; total time= 1.0s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=628; total time= 1.0s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=841; total time= 1.0s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=841; total time= 1.0s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=841; total time= 1.1s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=841; total time= 1.1s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=841; total time= 1.1s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=505; total time= 0.5s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=505; total time= 0.6s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=505; total time= 0.4s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=505; total time= 0.5s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=505; total time= 0.6s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=325; total time= 0.5s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=325; total time= 0.6s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=325; total time= 0.6s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=325; total time= 0.6s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=325; total time= 0.5s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=961; total time= 1.1s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=961; total time= 1.2s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=961; total time= 1.2s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=961; total time= 1.2s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=961; total time= 1.2s [CV] END bootstrap=False, max_depth=205, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=763; total time= 0.8s [CV] END bootstrap=False, max_depth=205, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=763; total time= 0.9s [CV] END bootstrap=False, max_depth=205, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=763; total time= 0.8s [CV] END bootstrap=False, max_depth=205, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=763; total time= 0.7s [CV] END bootstrap=False, max_depth=205, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=763; total time= 0.8s [CV] END bootstrap=True, max_depth=40, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=94; total time= 0.1s [CV] END bootstrap=True, max_depth=40, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=94; total time= 0.1s [CV] END bootstrap=True, max_depth=40, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=94; total time= 0.1s [CV] END bootstrap=True, max_depth=40, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=94; total time= 0.1s [CV] END bootstrap=True, max_depth=40, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=94; total time= 0.1s [CV] END bootstrap=False, max_depth=15, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=856; total time= 1.0s [CV] END bootstrap=False, max_depth=15, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=856; total time= 1.1s [CV] END bootstrap=False, max_depth=15, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=856; total time= 1.2s [CV] END bootstrap=False, max_depth=15, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=856; total time= 1.1s [CV] END bootstrap=False, max_depth=15, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=856; total time= 1.1s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=853; total time= 1.1s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=853; total time= 1.1s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=853; total time= 1.1s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=853; total time= 1.2s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=853; total time= 1.1s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=73; total time= 0.1s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=73; total time= 0.1s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=73; total time= 0.1s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=73; total time= 0.1s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=73; total time= 0.1s [CV] END bootstrap=True, max_depth=20, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=448; total time= 0.6s [CV] END bootstrap=True, max_depth=20, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=448; total time= 0.6s [CV] END bootstrap=True, max_depth=20, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=448; total time= 0.7s [CV] END bootstrap=True, max_depth=20, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=448; total time= 0.7s [CV] END bootstrap=True, max_depth=20, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=448; total time= 0.6s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=661; total time= 0.9s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=661; total time= 0.9s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=661; total time= 0.9s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=661; total time= 1.0s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=661; total time= 0.9s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=727; total time= 1.1s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=727; total time= 1.0s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=727; total time= 0.9s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=727; total time= 1.0s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=727; total time= 1.0s [CV] END bootstrap=True, max_depth=220, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=76; total time= 0.1s [CV] END bootstrap=True, max_depth=220, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=76; total time= 0.2s [CV] END bootstrap=True, max_depth=220, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=76; total time= 0.1s [CV] END bootstrap=True, max_depth=220, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=76; total time= 0.1s [CV] END bootstrap=True, max_depth=220, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=76; total time= 0.1s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=202; total time= 0.3s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=202; total time= 0.3s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=202; total time= 0.3s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=202; total time= 0.3s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=202; total time= 0.2s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=22; total time= 0.0s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=22; total time= 0.0s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=22; total time= 0.0s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=22; total time= 0.0s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=22; total time= 0.0s [CV] END bootstrap=False, max_depth=20, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=538; total time= 0.7s [CV] END bootstrap=False, max_depth=20, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=538; total time= 0.8s [CV] END bootstrap=False, max_depth=20, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=538; total time= 0.6s [CV] END bootstrap=False, max_depth=20, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=538; total time= 0.7s [CV] END bootstrap=False, max_depth=20, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=538; total time= 0.7s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=766; total time= 0.8s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=766; total time= 0.8s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=766; total time= 0.8s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=766; total time= 0.7s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=766; total time= 0.8s [CV] END bootstrap=False, max_depth=60, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=682; total time= 0.9s [CV] END bootstrap=False, max_depth=60, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=682; total time= 1.0s [CV] END bootstrap=False, max_depth=60, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=682; total time= 0.8s [CV] END bootstrap=False, max_depth=60, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=682; total time= 0.8s [CV] END bootstrap=False, max_depth=60, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=682; total time= 0.9s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=808; total time= 1.0s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=808; total time= 0.9s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=808; total time= 0.9s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=808; total time= 1.0s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=808; total time= 1.0s [CV] END bootstrap=True, max_depth=210, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=310; total time= 0.5s [CV] END bootstrap=True, max_depth=210, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=310; total time= 0.5s [CV] END bootstrap=True, max_depth=210, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=310; total time= 0.5s [CV] END bootstrap=True, max_depth=210, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=310; total time= 0.4s [CV] END bootstrap=True, max_depth=210, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=310; total time= 0.5s [CV] END bootstrap=False, max_depth=160, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=193; total time= 0.2s [CV] END bootstrap=False, max_depth=160, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=193; total time= 0.2s [CV] END bootstrap=False, max_depth=160, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=193; total time= 0.3s [CV] END bootstrap=False, max_depth=160, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=193; total time= 0.2s [CV] END bootstrap=False, max_depth=160, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=193; total time= 0.3s [CV] END bootstrap=False, max_depth=25, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=793; total time= 1.0s [CV] END bootstrap=False, max_depth=25, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=793; total time= 0.9s [CV] END bootstrap=False, max_depth=25, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=793; total time= 0.9s [CV] END bootstrap=False, max_depth=25, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=793; total time= 0.9s [CV] END bootstrap=False, max_depth=25, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=793; total time= 1.0s [CV] END bootstrap=False, max_depth=225, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=757; total time= 0.7s [CV] END bootstrap=False, max_depth=225, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=757; total time= 0.7s [CV] END bootstrap=False, max_depth=225, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=757; total time= 0.8s [CV] END bootstrap=False, max_depth=225, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=757; total time= 0.8s [CV] END bootstrap=False, max_depth=225, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=757; total time= 0.7s [CV] END bootstrap=True, max_depth=225, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=748; total time= 0.8s [CV] END bootstrap=True, max_depth=225, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=748; total time= 0.9s [CV] END bootstrap=True, max_depth=225, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=748; total time= 0.9s [CV] END bootstrap=True, max_depth=225, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=748; total time= 1.0s [CV] END bootstrap=True, max_depth=225, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=748; total time= 1.0s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=907; total time= 1.3s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=907; total time= 1.3s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=907; total time= 1.3s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=907; total time= 1.3s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=907; total time= 1.3s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=928; total time= 1.0s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=928; total time= 1.1s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=928; total time= 1.1s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=928; total time= 1.1s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=928; total time= 1.0s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=820; total time= 0.9s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=820; total time= 1.0s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=820; total time= 1.0s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=820; total time= 1.0s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=820; total time= 1.0s [CV] END bootstrap=False, max_depth=15, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=811; total time= 0.8s [CV] END bootstrap=False, max_depth=15, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=811; total time= 1.0s [CV] END bootstrap=False, max_depth=15, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=811; total time= 1.0s [CV] END bootstrap=False, max_depth=15, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=811; total time= 0.9s [CV] END bootstrap=False, max_depth=15, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=811; total time= 1.0s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=613; total time= 0.9s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=613; total time= 0.9s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=613; total time= 0.9s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=613; total time= 0.8s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=613; total time= 0.9s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=337; total time= 0.6s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=337; total time= 0.6s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=337; total time= 0.6s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=337; total time= 0.6s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=337; total time= 0.6s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=988; total time= 1.2s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=988; total time= 1.2s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=988; total time= 1.3s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=988; total time= 1.2s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=988; total time= 1.5s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=148; total time= 0.3s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=148; total time= 0.3s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=148; total time= 0.2s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=148; total time= 0.2s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=148; total time= 0.2s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=541; total time= 0.7s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=541; total time= 0.6s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=541; total time= 0.6s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=541; total time= 0.7s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=541; total time= 0.6s [CV] END bootstrap=False, max_depth=160, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=43; total time= 0.0s [CV] END bootstrap=False, max_depth=160, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=43; total time= 0.0s [CV] END bootstrap=False, max_depth=160, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=43; total time= 0.0s [CV] END bootstrap=False, max_depth=160, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=43; total time= 0.1s [CV] END bootstrap=False, max_depth=160, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=43; total time= 0.0s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=931; total time= 1.0s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=931; total time= 1.1s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=931; total time= 1.1s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=931; total time= 1.1s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=931; total time= 1.2s [CV] END bootstrap=True, max_depth=255, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=763; total time= 0.6s [CV] END bootstrap=True, max_depth=255, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=763; total time= 0.7s [CV] END bootstrap=True, max_depth=255, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=763; total time= 0.8s [CV] END bootstrap=True, max_depth=255, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=763; total time= 0.8s [CV] END bootstrap=True, max_depth=255, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=763; total time= 0.7s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=328; total time= 0.4s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=328; total time= 0.5s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=328; total time= 0.5s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=328; total time= 0.5s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=328; total time= 0.6s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=502; total time= 0.8s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=502; total time= 0.9s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=502; total time= 0.4s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=502; total time= 0.6s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=502; total time= 0.5s [CV] END bootstrap=True, max_depth=20, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=16; total time= 0.0s [CV] END bootstrap=True, max_depth=20, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=16; total time= 0.0s [CV] END bootstrap=True, max_depth=20, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=16; total time= 0.0s [CV] END bootstrap=True, max_depth=20, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=16; total time= 0.0s [CV] END bootstrap=True, max_depth=20, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=16; total time= 0.0s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=61; total time= 0.1s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=61; total time= 0.0s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=61; total time= 0.1s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=61; total time= 0.0s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=61; total time= 0.1s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=412; total time= 0.5s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=412; total time= 0.5s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=412; total time= 0.5s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=412; total time= 0.5s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=412; total time= 0.6s [CV] END bootstrap=False, max_depth=75, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=73; total time= 0.1s [CV] END bootstrap=False, max_depth=75, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=73; total time= 0.1s [CV] END bootstrap=False, max_depth=75, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=73; total time= 0.1s [CV] END bootstrap=False, max_depth=75, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=73; total time= 0.1s [CV] END bootstrap=False, max_depth=75, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=73; total time= 0.1s [CV] END bootstrap=False, max_depth=210, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=55; total time= 0.1s [CV] END bootstrap=False, max_depth=210, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=55; total time= 0.1s [CV] END bootstrap=False, max_depth=210, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=55; total time= 0.1s [CV] END bootstrap=False, max_depth=210, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=55; total time= 0.1s [CV] END bootstrap=False, max_depth=210, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=55; total time= 0.1s [CV] END bootstrap=True, max_depth=80, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=319; total time= 0.5s [CV] END bootstrap=True, max_depth=80, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=319; total time= 0.5s [CV] END bootstrap=True, max_depth=80, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=319; total time= 0.5s [CV] END bootstrap=True, max_depth=80, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=319; total time= 0.6s [CV] END bootstrap=True, max_depth=80, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=319; total time= 0.5s [CV] END bootstrap=True, max_depth=215, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=958; total time= 1.1s [CV] END bootstrap=True, max_depth=215, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=958; total time= 1.1s [CV] END bootstrap=True, max_depth=215, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=958; total time= 1.1s [CV] END bootstrap=True, max_depth=215, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=958; total time= 1.1s [CV] END bootstrap=True, max_depth=215, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=958; total time= 1.1s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=448; total time= 0.7s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=448; total time= 0.6s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=448; total time= 0.8s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=448; total time= 0.7s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=448; total time= 0.5s [CV] END bootstrap=False, max_depth=255, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=268; total time= 0.4s [CV] END bootstrap=False, max_depth=255, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=268; total time= 0.4s [CV] END bootstrap=False, max_depth=255, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=268; total time= 0.4s [CV] END bootstrap=False, max_depth=255, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=268; total time= 0.3s [CV] END bootstrap=False, max_depth=255, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=268; total time= 0.3s [CV] END bootstrap=False, max_depth=190, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=235; total time= 0.4s [CV] END bootstrap=False, max_depth=190, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=235; total time= 0.4s [CV] END bootstrap=False, max_depth=190, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=235; total time= 0.3s [CV] END bootstrap=False, max_depth=190, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=235; total time= 0.4s [CV] END bootstrap=False, max_depth=190, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=235; total time= 0.4s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=730; total time= 1.1s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=730; total time= 1.1s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=730; total time= 1.0s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=730; total time= 1.0s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=730; total time= 1.0s [CV] END bootstrap=True, max_depth=35, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=406; total time= 0.5s [CV] END bootstrap=True, max_depth=35, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=406; total time= 0.5s [CV] END bootstrap=True, max_depth=35, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=406; total time= 0.6s [CV] END bootstrap=True, max_depth=35, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=406; total time= 0.6s [CV] END bootstrap=True, max_depth=35, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=406; total time= 0.5s [CV] END bootstrap=True, max_depth=250, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=997; total time= 1.2s [CV] END bootstrap=True, max_depth=250, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=997; total time= 1.2s [CV] END bootstrap=True, max_depth=250, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=997; total time= 1.3s [CV] END bootstrap=True, max_depth=250, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=997; total time= 1.2s [CV] END bootstrap=True, max_depth=250, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=997; total time= 1.2s [CV] END bootstrap=False, max_depth=195, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=436; total time= 0.6s [CV] END bootstrap=False, max_depth=195, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=436; total time= 0.6s [CV] END bootstrap=False, max_depth=195, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=436; total time= 0.6s [CV] END bootstrap=False, max_depth=195, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=436; total time= 0.6s [CV] END bootstrap=False, max_depth=195, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=436; total time= 0.6s [CV] END bootstrap=False, max_depth=180, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=805; total time= 0.9s [CV] END bootstrap=False, max_depth=180, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=805; total time= 0.9s [CV] END bootstrap=False, max_depth=180, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=805; total time= 0.9s [CV] END bootstrap=False, max_depth=180, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=805; total time= 1.0s [CV] END bootstrap=False, max_depth=180, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=805; total time= 0.9s [CV] END bootstrap=True, max_depth=30, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=190; total time= 0.2s [CV] END bootstrap=True, max_depth=30, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=190; total time= 0.2s [CV] END bootstrap=True, max_depth=30, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=190; total time= 0.3s [CV] END bootstrap=True, max_depth=30, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=190; total time= 0.2s [CV] END bootstrap=True, max_depth=30, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=190; total time= 0.2s [CV] END bootstrap=False, max_depth=230, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=964; total time= 1.1s [CV] END bootstrap=False, max_depth=230, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=964; total time= 1.2s [CV] END bootstrap=False, max_depth=230, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=964; total time= 1.1s [CV] END bootstrap=False, max_depth=230, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=964; total time= 1.1s [CV] END bootstrap=False, max_depth=230, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=964; total time= 1.2s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=445; total time= 0.7s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=445; total time= 0.6s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=445; total time= 0.7s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=445; total time= 0.7s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=445; total time= 0.6s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=625; total time= 0.8s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=625; total time= 0.9s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=625; total time= 0.9s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=625; total time= 0.9s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=625; total time= 0.9s [CV] END bootstrap=True, max_depth=60, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=214; total time= 0.3s [CV] END bootstrap=True, max_depth=60, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=214; total time= 0.3s [CV] END bootstrap=True, max_depth=60, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=214; total time= 0.3s [CV] END bootstrap=True, max_depth=60, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=214; total time= 0.3s [CV] END bootstrap=True, max_depth=60, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=214; total time= 0.3s [CV] END bootstrap=True, max_depth=10, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=673; total time= 0.9s [CV] END bootstrap=True, max_depth=10, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=673; total time= 1.1s [CV] END bootstrap=True, max_depth=10, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=673; total time= 0.9s [CV] END bootstrap=True, max_depth=10, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=673; total time= 0.8s [CV] END bootstrap=True, max_depth=10, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=673; total time= 0.8s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=82; total time= 0.1s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=82; total time= 0.1s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=82; total time= 0.2s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=82; total time= 0.1s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=82; total time= 0.1s [CV] END bootstrap=False, max_depth=45, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=154; total time= 0.3s [CV] END bootstrap=False, max_depth=45, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=154; total time= 0.3s [CV] END bootstrap=False, max_depth=45, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=154; total time= 0.2s [CV] END bootstrap=False, max_depth=45, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=154; total time= 0.3s [CV] END bootstrap=False, max_depth=45, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=154; total time= 0.3s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=763; total time= 0.8s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=763; total time= 0.8s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=763; total time= 0.7s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=763; total time= 0.9s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=763; total time= 0.8s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=73; total time= 0.1s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=73; total time= 0.1s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=73; total time= 0.1s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=73; total time= 0.1s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=73; total time= 0.1s [CV] END bootstrap=False, max_depth=150, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=181; total time= 0.2s [CV] END bootstrap=False, max_depth=150, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=181; total time= 0.3s [CV] END bootstrap=False, max_depth=150, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=181; total time= 0.2s [CV] END bootstrap=False, max_depth=150, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=181; total time= 0.2s [CV] END bootstrap=False, max_depth=150, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=181; total time= 0.2s [CV] END bootstrap=False, max_depth=145, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=145; total time= 0.3s [CV] END bootstrap=False, max_depth=145, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=145; total time= 0.2s [CV] END bootstrap=False, max_depth=145, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=145; total time= 0.3s [CV] END bootstrap=False, max_depth=145, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=145; total time= 0.3s [CV] END bootstrap=False, max_depth=145, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=145; total time= 0.3s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=697; total time= 0.9s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=697; total time= 1.0s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=697; total time= 1.0s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=697; total time= 1.0s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=697; total time= 0.9s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=610; total time= 0.9s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=610; total time= 0.9s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=610; total time= 0.8s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=610; total time= 0.7s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=610; total time= 0.9s [CV] END bootstrap=True, max_depth=135, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=565; total time= 0.7s [CV] END bootstrap=True, max_depth=135, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=565; total time= 0.8s [CV] END bootstrap=True, max_depth=135, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=565; total time= 0.8s [CV] END bootstrap=True, max_depth=135, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=565; total time= 0.8s [CV] END bootstrap=True, max_depth=135, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=565; total time= 0.7s [CV] END bootstrap=False, max_depth=60, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=352; total time= 0.7s [CV] END bootstrap=False, max_depth=60, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=352; total time= 0.6s [CV] END bootstrap=False, max_depth=60, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=352; total time= 0.6s [CV] END bootstrap=False, max_depth=60, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=352; total time= 0.6s [CV] END bootstrap=False, max_depth=60, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=352; total time= 0.6s [CV] END bootstrap=True, max_depth=190, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=337; total time= 0.6s [CV] END bootstrap=True, max_depth=190, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=337; total time= 0.6s [CV] END bootstrap=True, max_depth=190, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=337; total time= 0.6s [CV] END bootstrap=True, max_depth=190, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=337; total time= 0.6s [CV] END bootstrap=True, max_depth=190, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=337; total time= 0.6s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=760; total time= 0.8s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=760; total time= 0.8s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=760; total time= 0.7s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=760; total time= 0.8s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=760; total time= 0.8s [CV] END bootstrap=True, max_depth=150, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=481; total time= 0.8s [CV] END bootstrap=True, max_depth=150, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=481; total time= 0.8s [CV] END bootstrap=True, max_depth=150, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=481; total time= 0.6s [CV] END bootstrap=True, max_depth=150, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=481; total time= 0.8s [CV] END bootstrap=True, max_depth=150, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=481; total time= 0.8s [CV] END bootstrap=True, max_depth=200, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=1000; total time= 1.0s [CV] END bootstrap=True, max_depth=200, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=1000; total time= 1.2s [CV] END bootstrap=True, max_depth=200, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=1000; total time= 1.2s [CV] END bootstrap=True, max_depth=200, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=1000; total time= 1.1s [CV] END bootstrap=True, max_depth=200, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=1000; total time= 1.2s [CV] END bootstrap=True, max_depth=180, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=934; total time= 1.1s [CV] END bootstrap=True, max_depth=180, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=934; total time= 1.1s [CV] END bootstrap=True, max_depth=180, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=934; total time= 1.1s [CV] END bootstrap=True, max_depth=180, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=934; total time= 1.2s [CV] END bootstrap=True, max_depth=180, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=934; total time= 1.1s [CV] END bootstrap=True, max_depth=200, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=613; total time= 0.9s [CV] END bootstrap=True, max_depth=200, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=613; total time= 0.9s [CV] END bootstrap=True, max_depth=200, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=613; total time= 1.0s [CV] END bootstrap=True, max_depth=200, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=613; total time= 0.9s [CV] END bootstrap=True, max_depth=200, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=613; total time= 0.9s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=643; total time= 1.0s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=643; total time= 1.0s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=643; total time= 0.9s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=643; total time= 1.0s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=643; total time= 1.0s [CV] END bootstrap=False, max_depth=40, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=946; total time= 1.2s [CV] END bootstrap=False, max_depth=40, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=946; total time= 1.3s [CV] END bootstrap=False, max_depth=40, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=946; total time= 1.1s [CV] END bootstrap=False, max_depth=40, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=946; total time= 1.2s [CV] END bootstrap=False, max_depth=40, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=946; total time= 1.1s [CV] END bootstrap=True, max_depth=245, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=616; total time= 0.9s [CV] END bootstrap=True, max_depth=245, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=616; total time= 1.0s [CV] END bootstrap=True, max_depth=245, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=616; total time= 0.9s [CV] END bootstrap=True, max_depth=245, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=616; total time= 0.9s [CV] END bootstrap=True, max_depth=245, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=616; total time= 0.9s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=262; total time= 0.5s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=262; total time= 0.3s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=262; total time= 0.3s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=262; total time= 0.3s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=262; total time= 0.4s [CV] END bootstrap=True, max_depth=45, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=55; total time= 0.1s [CV] END bootstrap=True, max_depth=45, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=55; total time= 0.1s [CV] END bootstrap=True, max_depth=45, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=55; total time= 0.0s [CV] END bootstrap=True, max_depth=45, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=55; total time= 0.0s [CV] END bootstrap=True, max_depth=45, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=55; total time= 0.1s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=787; total time= 0.8s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=787; total time= 0.8s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=787; total time= 0.9s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=787; total time= 0.8s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=787; total time= 0.9s [CV] END bootstrap=True, max_depth=230, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=382; total time= 0.5s [CV] END bootstrap=True, max_depth=230, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=382; total time= 0.4s [CV] END bootstrap=True, max_depth=230, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=382; total time= 0.5s [CV] END bootstrap=True, max_depth=230, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=382; total time= 0.5s [CV] END bootstrap=True, max_depth=230, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=382; total time= 0.5s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=919; total time= 1.0s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=919; total time= 1.1s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=919; total time= 1.2s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=919; total time= 1.3s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=919; total time= 1.0s [CV] END bootstrap=True, max_depth=175, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=541; total time= 0.6s [CV] END bootstrap=True, max_depth=175, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=541; total time= 0.5s [CV] END bootstrap=True, max_depth=175, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=541; total time= 0.7s [CV] END bootstrap=True, max_depth=175, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=541; total time= 0.6s [CV] END bootstrap=True, max_depth=175, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=541; total time= 0.6s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=469; total time= 0.6s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=469; total time= 0.8s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=469; total time= 0.8s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=469; total time= 0.7s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=469; total time= 0.7s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=451; total time= 0.7s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=451; total time= 0.6s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=451; total time= 0.7s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=451; total time= 0.6s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=451; total time= 0.6s [CV] END bootstrap=False, max_depth=85, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=667; total time= 0.7s [CV] END bootstrap=False, max_depth=85, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=667; total time= 0.9s [CV] END bootstrap=False, max_depth=85, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=667; total time= 0.9s [CV] END bootstrap=False, max_depth=85, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=667; total time= 0.9s [CV] END bootstrap=False, max_depth=85, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=667; total time= 0.9s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=25; total time= 0.0s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=25; total time= 0.0s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=25; total time= 0.0s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=25; total time= 0.0s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=25; total time= 0.0s [CV] END bootstrap=True, max_depth=210, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=100; total time= 0.1s [CV] END bootstrap=True, max_depth=210, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=100; total time= 0.2s [CV] END bootstrap=True, max_depth=210, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=100; total time= 0.2s [CV] END bootstrap=True, max_depth=210, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=100; total time= 0.1s [CV] END bootstrap=True, max_depth=210, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=100; total time= 0.1s [CV] END bootstrap=False, max_depth=235, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=655; total time= 1.0s [CV] END bootstrap=False, max_depth=235, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=655; total time= 1.0s [CV] END bootstrap=False, max_depth=235, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=655; total time= 1.0s [CV] END bootstrap=False, max_depth=235, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=655; total time= 1.0s [CV] END bootstrap=False, max_depth=235, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=655; total time= 1.1s [CV] END bootstrap=False, max_depth=60, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=238; total time= 0.4s [CV] END bootstrap=False, max_depth=60, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=238; total time= 0.4s [CV] END bootstrap=False, max_depth=60, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=238; total time= 0.4s [CV] END bootstrap=False, max_depth=60, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=238; total time= 0.4s [CV] END bootstrap=False, max_depth=60, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=238; total time= 0.4s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=331; total time= 0.5s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=331; total time= 0.6s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=331; total time= 0.6s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=331; total time= 0.6s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=331; total time= 0.6s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=13; total time= 0.0s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=13; total time= 0.0s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=13; total time= 0.0s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=13; total time= 0.0s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=13; total time= 0.0s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=817; total time= 0.9s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=817; total time= 1.0s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=817; total time= 1.0s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=817; total time= 0.9s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=817; total time= 0.9s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=61; total time= 0.1s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=61; total time= 0.1s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=61; total time= 0.1s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=61; total time= 0.1s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=61; total time= 0.0s [CV] END bootstrap=False, max_depth=65, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=370; total time= 0.8s [CV] END bootstrap=False, max_depth=65, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=370; total time= 0.4s [CV] END bootstrap=False, max_depth=65, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=370; total time= 0.4s [CV] END bootstrap=False, max_depth=65, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=370; total time= 0.4s [CV] END bootstrap=False, max_depth=65, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=370; total time= 0.5s [CV] END bootstrap=False, max_depth=195, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=802; total time= 0.9s [CV] END bootstrap=False, max_depth=195, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=802; total time= 0.9s [CV] END bootstrap=False, max_depth=195, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=802; total time= 0.9s [CV] END bootstrap=False, max_depth=195, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=802; total time= 0.9s [CV] END bootstrap=False, max_depth=195, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=802; total time= 0.9s [CV] END bootstrap=True, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=325; total time= 0.6s [CV] END bootstrap=True, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=325; total time= 0.5s [CV] END bootstrap=True, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=325; total time= 0.5s [CV] END bootstrap=True, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=325; total time= 0.5s [CV] END bootstrap=True, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=325; total time= 0.6s [CV] END bootstrap=False, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=622; total time= 0.8s [CV] END bootstrap=False, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=622; total time= 0.9s [CV] END bootstrap=False, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=622; total time= 0.8s [CV] END bootstrap=False, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=622; total time= 0.9s [CV] END bootstrap=False, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=622; total time= 0.9s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=349; total time= 0.6s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=349; total time= 0.5s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=349; total time= 0.6s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=349; total time= 0.7s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=349; total time= 0.5s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=688; total time= 0.8s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=688; total time= 0.9s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=688; total time= 0.9s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=688; total time= 1.0s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=688; total time= 0.9s [CV] END bootstrap=False, max_depth=235, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=784; total time= 0.9s [CV] END bootstrap=False, max_depth=235, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=784; total time= 0.8s [CV] END bootstrap=False, max_depth=235, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=784; total time= 0.9s [CV] END bootstrap=False, max_depth=235, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=784; total time= 0.9s [CV] END bootstrap=False, max_depth=235, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=784; total time= 0.8s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=469; total time= 0.7s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=469; total time= 0.7s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=469; total time= 0.7s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=469; total time= 0.7s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=469; total time= 0.7s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=313; total time= 0.5s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=313; total time= 0.4s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=313; total time= 0.5s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=313; total time= 0.5s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=313; total time= 0.5s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=445; total time= 0.6s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=445; total time= 0.9s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=445; total time= 0.5s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=445; total time= 0.6s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=445; total time= 0.7s [CV] END bootstrap=False, max_depth=200, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=943; total time= 1.1s [CV] END bootstrap=False, max_depth=200, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=943; total time= 1.1s [CV] END bootstrap=False, max_depth=200, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=943; total time= 1.1s [CV] END bootstrap=False, max_depth=200, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=943; total time= 1.2s [CV] END bootstrap=False, max_depth=200, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=943; total time= 1.2s [CV] END bootstrap=False, max_depth=90, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=109; total time= 0.2s [CV] END bootstrap=False, max_depth=90, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=109; total time= 0.2s [CV] END bootstrap=False, max_depth=90, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=109; total time= 0.1s [CV] END bootstrap=False, max_depth=90, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=109; total time= 0.2s [CV] END bootstrap=False, max_depth=90, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=109; total time= 0.2s [CV] END bootstrap=False, max_depth=205, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=583; total time= 0.8s [CV] END bootstrap=False, max_depth=205, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=583; total time= 0.8s [CV] END bootstrap=False, max_depth=205, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=583; total time= 0.7s [CV] END bootstrap=False, max_depth=205, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=583; total time= 0.8s [CV] END bootstrap=False, max_depth=205, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=583; total time= 0.8s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=130; total time= 0.2s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=130; total time= 0.2s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=130; total time= 0.2s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=130; total time= 0.2s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=130; total time= 0.2s [CV] END bootstrap=True, max_depth=245, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=412; total time= 0.5s [CV] END bootstrap=True, max_depth=245, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=412; total time= 0.6s [CV] END bootstrap=True, max_depth=245, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=412; total time= 0.6s [CV] END bootstrap=True, max_depth=245, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=412; total time= 0.5s [CV] END bootstrap=True, max_depth=245, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=412; total time= 0.6s [CV] END bootstrap=True, max_depth=240, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=106; total time= 0.2s [CV] END bootstrap=True, max_depth=240, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=106; total time= 0.2s [CV] END bootstrap=True, max_depth=240, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=106; total time= 0.2s [CV] END bootstrap=True, max_depth=240, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=106; total time= 0.2s [CV] END bootstrap=True, max_depth=240, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=106; total time= 0.2s [CV] END bootstrap=True, max_depth=225, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=127; total time= 0.2s [CV] END bootstrap=True, max_depth=225, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=127; total time= 0.2s [CV] END bootstrap=True, max_depth=225, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=127; total time= 0.2s [CV] END bootstrap=True, max_depth=225, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=127; total time= 0.2s [CV] END bootstrap=True, max_depth=225, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=127; total time= 0.2s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=832; total time= 1.0s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=832; total time= 1.1s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=832; total time= 1.0s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=832; total time= 1.2s [CV] END bootstrap=False, max_depth=95, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=832; total time= 1.0s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=415; total time= 0.5s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=415; total time= 0.5s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=415; total time= 0.5s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=415; total time= 0.5s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=415; total time= 0.6s [CV] END bootstrap=True, max_depth=175, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=682; total time= 1.0s [CV] END bootstrap=True, max_depth=175, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=682; total time= 0.9s [CV] END bootstrap=True, max_depth=175, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=682; total time= 0.8s [CV] END bootstrap=True, max_depth=175, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=682; total time= 0.9s [CV] END bootstrap=True, max_depth=175, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=682; total time= 0.9s [CV] END bootstrap=False, max_depth=70, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=931; total time= 1.1s [CV] END bootstrap=False, max_depth=70, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=931; total time= 1.1s [CV] END bootstrap=False, max_depth=70, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=931; total time= 1.2s [CV] END bootstrap=False, max_depth=70, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=931; total time= 1.0s [CV] END bootstrap=False, max_depth=70, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=931; total time= 1.2s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=685; total time= 0.9s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=685; total time= 0.9s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=685; total time= 0.9s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=685; total time= 0.8s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=685; total time= 1.0s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=508; total time= 0.5s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=508; total time= 0.5s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=508; total time= 0.6s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=508; total time= 0.5s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=508; total time= 0.6s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=673; total time= 0.9s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=673; total time= 1.0s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=673; total time= 0.9s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=673; total time= 0.9s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=673; total time= 1.0s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=577; total time= 0.7s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=577; total time= 0.8s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=577; total time= 0.8s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=577; total time= 0.7s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=577; total time= 0.8s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=550; total time= 0.8s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=550; total time= 0.7s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=550; total time= 0.8s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=550; total time= 0.7s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=550; total time= 0.7s [CV] END bootstrap=False, max_depth=140, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=181; total time= 0.3s [CV] END bootstrap=False, max_depth=140, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=181; total time= 0.2s [CV] END bootstrap=False, max_depth=140, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=181; total time= 0.2s [CV] END bootstrap=False, max_depth=140, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=181; total time= 0.2s [CV] END bootstrap=False, max_depth=140, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=181; total time= 0.4s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=754; total time= 0.9s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=754; total time= 1.0s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=754; total time= 0.9s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=754; total time= 0.7s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=754; total time= 0.7s [CV] END bootstrap=False, max_depth=170, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=859; total time= 1.2s [CV] END bootstrap=False, max_depth=170, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=859; total time= 1.0s [CV] END bootstrap=False, max_depth=170, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=859; total time= 1.2s [CV] END bootstrap=False, max_depth=170, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=859; total time= 1.1s [CV] END bootstrap=False, max_depth=170, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=859; total time= 1.0s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=565; total time= 0.8s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=565; total time= 0.9s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=565; total time= 0.7s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=565; total time= 0.7s [CV] END bootstrap=True, max_depth=120, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=565; total time= 0.7s [CV] END bootstrap=True, max_depth=95, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=736; total time= 1.0s [CV] END bootstrap=True, max_depth=95, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=736; total time= 0.9s [CV] END bootstrap=True, max_depth=95, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=736; total time= 1.0s [CV] END bootstrap=True, max_depth=95, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=736; total time= 1.0s [CV] END bootstrap=True, max_depth=95, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=736; total time= 0.9s [CV] END bootstrap=True, max_depth=185, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=457; total time= 0.7s [CV] END bootstrap=True, max_depth=185, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=457; total time= 0.7s [CV] END bootstrap=True, max_depth=185, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=457; total time= 0.7s [CV] END bootstrap=True, max_depth=185, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=457; total time= 0.7s [CV] END bootstrap=True, max_depth=185, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=457; total time= 0.7s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=904; total time= 1.2s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=904; total time= 1.2s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=904; total time= 1.1s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=904; total time= 1.3s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=904; total time= 1.3s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=679; total time= 0.9s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=679; total time= 0.8s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=679; total time= 0.8s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=679; total time= 0.9s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=679; total time= 0.9s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=895; total time= 1.2s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=895; total time= 1.2s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=895; total time= 1.3s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=895; total time= 1.2s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=895; total time= 1.2s [CV] END bootstrap=False, max_depth=215, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=520; total time= 0.6s [CV] END bootstrap=False, max_depth=215, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=520; total time= 0.6s [CV] END bootstrap=False, max_depth=215, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=520; total time= 0.6s [CV] END bootstrap=False, max_depth=215, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=520; total time= 0.6s [CV] END bootstrap=False, max_depth=215, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=520; total time= 0.6s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=814; total time= 0.9s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=814; total time= 0.9s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=814; total time= 0.9s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=814; total time= 1.0s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=814; total time= 0.8s [CV] END bootstrap=False, max_depth=185, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=124; total time= 0.1s [CV] END bootstrap=False, max_depth=185, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=124; total time= 0.1s [CV] END bootstrap=False, max_depth=185, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=124; total time= 0.2s [CV] END bootstrap=False, max_depth=185, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=124; total time= 0.2s [CV] END bootstrap=False, max_depth=185, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=124; total time= 0.1s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=343; total time= 0.6s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=343; total time= 0.6s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=343; total time= 0.6s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=343; total time= 0.6s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=343; total time= 0.5s [CV] END bootstrap=False, max_depth=195, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=883; total time= 1.2s [CV] END bootstrap=False, max_depth=195, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=883; total time= 1.2s [CV] END bootstrap=False, max_depth=195, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=883; total time= 1.3s [CV] END bootstrap=False, max_depth=195, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=883; total time= 1.3s [CV] END bootstrap=False, max_depth=195, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=883; total time= 1.1s [CV] END bootstrap=True, max_depth=30, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=418; total time= 0.6s [CV] END bootstrap=True, max_depth=30, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=418; total time= 0.5s [CV] END bootstrap=True, max_depth=30, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=418; total time= 0.6s [CV] END bootstrap=True, max_depth=30, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=418; total time= 0.5s [CV] END bootstrap=True, max_depth=30, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=418; total time= 0.5s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=205; total time= 0.3s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=205; total time= 0.3s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=205; total time= 0.3s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=205; total time= 0.2s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=205; total time= 0.2s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=742; total time= 0.8s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=742; total time= 1.0s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=742; total time= 0.9s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=742; total time= 1.0s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=742; total time= 1.0s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=502; total time= 0.5s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=502; total time= 0.4s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=502; total time= 0.6s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=502; total time= 0.8s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=502; total time= 0.4s [CV] END bootstrap=False, max_depth=25, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=133; total time= 0.3s [CV] END bootstrap=False, max_depth=25, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=133; total time= 0.2s [CV] END bootstrap=False, max_depth=25, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=133; total time= 0.2s [CV] END bootstrap=False, max_depth=25, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=133; total time= 0.2s [CV] END bootstrap=False, max_depth=25, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=133; total time= 0.2s [CV] END bootstrap=False, max_depth=115, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=967; total time= 1.1s [CV] END bootstrap=False, max_depth=115, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=967; total time= 1.3s [CV] END bootstrap=False, max_depth=115, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=967; total time= 1.2s [CV] END bootstrap=False, max_depth=115, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=967; total time= 1.1s [CV] END bootstrap=False, max_depth=115, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=967; total time= 1.2s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=739; total time= 0.9s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=739; total time= 0.9s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=739; total time= 0.9s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=739; total time= 0.7s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=739; total time= 0.7s [CV] END bootstrap=True, max_depth=50, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=988; total time= 1.6s [CV] END bootstrap=True, max_depth=50, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=988; total time= 1.2s [CV] END bootstrap=True, max_depth=50, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=988; total time= 1.2s [CV] END bootstrap=True, max_depth=50, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=988; total time= 1.2s [CV] END bootstrap=True, max_depth=50, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=988; total time= 1.3s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=109; total time= 0.2s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=109; total time= 0.2s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=109; total time= 0.2s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=109; total time= 0.2s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=109; total time= 0.2s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=352; total time= 0.7s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=352; total time= 0.6s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=352; total time= 0.5s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=352; total time= 0.6s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=352; total time= 0.6s [CV] END bootstrap=True, max_depth=195, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=331; total time= 0.6s [CV] END bootstrap=True, max_depth=195, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=331; total time= 0.6s [CV] END bootstrap=True, max_depth=195, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=331; total time= 0.5s [CV] END bootstrap=True, max_depth=195, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=331; total time= 0.5s [CV] END bootstrap=True, max_depth=195, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=331; total time= 0.5s [CV] END bootstrap=True, max_depth=240, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=457; total time= 0.7s [CV] END bootstrap=True, max_depth=240, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=457; total time= 0.7s [CV] END bootstrap=True, max_depth=240, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=457; total time= 0.8s [CV] END bootstrap=True, max_depth=240, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=457; total time= 0.7s [CV] END bootstrap=True, max_depth=240, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=457; total time= 0.7s [CV] END bootstrap=False, max_depth=90, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=340; total time= 0.6s [CV] END bootstrap=False, max_depth=90, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=340; total time= 0.6s [CV] END bootstrap=False, max_depth=90, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=340; total time= 0.6s [CV] END bootstrap=False, max_depth=90, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=340; total time= 0.6s [CV] END bootstrap=False, max_depth=90, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=340; total time= 0.5s [CV] END bootstrap=True, max_depth=100, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=955; total time= 1.2s [CV] END bootstrap=True, max_depth=100, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=955; total time= 1.2s [CV] END bootstrap=True, max_depth=100, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=955; total time= 1.3s [CV] END bootstrap=True, max_depth=100, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=955; total time= 1.2s [CV] END bootstrap=True, max_depth=100, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=955; total time= 1.1s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=964; total time= 1.1s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=964; total time= 1.2s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=964; total time= 1.3s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=964; total time= 1.1s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=964; total time= 1.1s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=796; total time= 0.9s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=796; total time= 0.8s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=796; total time= 0.9s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=796; total time= 0.9s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=796; total time= 0.9s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=409; total time= 0.5s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=409; total time= 0.5s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=409; total time= 0.5s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=409; total time= 0.6s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=409; total time= 0.5s [CV] END bootstrap=False, max_depth=185, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=985; total time= 1.3s [CV] END bootstrap=False, max_depth=185, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=985; total time= 1.2s [CV] END bootstrap=False, max_depth=185, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=985; total time= 1.2s [CV] END bootstrap=False, max_depth=185, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=985; total time= 1.3s [CV] END bootstrap=False, max_depth=185, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=985; total time= 1.3s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=880; total time= 1.2s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=880; total time= 1.2s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=880; total time= 1.2s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=880; total time= 1.0s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=880; total time= 1.1s [CV] END bootstrap=True, max_depth=45, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=589; total time= 0.8s [CV] END bootstrap=True, max_depth=45, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=589; total time= 0.9s [CV] END bootstrap=True, max_depth=45, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=589; total time= 0.7s [CV] END bootstrap=True, max_depth=45, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=589; total time= 0.7s [CV] END bootstrap=True, max_depth=45, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=589; total time= 0.8s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=394; total time= 0.4s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=394; total time= 0.5s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=394; total time= 0.4s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=394; total time= 0.5s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=394; total time= 0.4s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=907; total time= 1.3s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=907; total time= 1.2s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=907; total time= 1.1s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=907; total time= 1.2s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=907; total time= 1.2s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=454; total time= 0.6s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=454; total time= 0.6s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=454; total time= 0.6s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=454; total time= 0.6s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=454; total time= 0.6s [CV] END bootstrap=False, max_depth=185, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=958; total time= 1.5s [CV] END bootstrap=False, max_depth=185, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=958; total time= 1.2s [CV] END bootstrap=False, max_depth=185, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=958; total time= 1.2s [CV] END bootstrap=False, max_depth=185, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=958; total time= 1.2s [CV] END bootstrap=False, max_depth=185, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=958; total time= 1.1s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=931; total time= 1.0s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=931; total time= 1.1s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=931; total time= 1.2s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=931; total time= 1.1s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=931; total time= 1.1s [CV] END bootstrap=True, max_depth=85, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=679; total time= 0.8s [CV] END bootstrap=True, max_depth=85, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=679; total time= 0.9s [CV] END bootstrap=True, max_depth=85, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=679; total time= 0.9s [CV] END bootstrap=True, max_depth=85, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=679; total time= 0.9s [CV] END bootstrap=True, max_depth=85, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=679; total time= 0.9s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=712; total time= 0.9s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=712; total time= 0.9s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=712; total time= 1.0s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=712; total time= 0.9s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=712; total time= 1.0s [CV] END bootstrap=True, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=202; total time= 0.2s [CV] END bootstrap=True, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=202; total time= 0.3s [CV] END bootstrap=True, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=202; total time= 0.2s [CV] END bootstrap=True, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=202; total time= 0.3s [CV] END bootstrap=True, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=202; total time= 0.3s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=538; total time= 0.7s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=538; total time= 0.7s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=538; total time= 0.7s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=538; total time= 0.6s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=538; total time= 0.7s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=985; total time= 1.2s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=985; total time= 1.2s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=985; total time= 1.2s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=985; total time= 1.2s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=985; total time= 1.1s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=91; total time= 0.1s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=91; total time= 0.2s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=91; total time= 0.1s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=91; total time= 0.1s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=91; total time= 0.2s [CV] END bootstrap=True, max_depth=125, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=394; total time= 0.5s [CV] END bootstrap=True, max_depth=125, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=394; total time= 0.4s [CV] END bootstrap=True, max_depth=125, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=394; total time= 0.5s [CV] END bootstrap=True, max_depth=125, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=394; total time= 0.4s [CV] END bootstrap=True, max_depth=125, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=394; total time= 0.5s [CV] END bootstrap=False, max_depth=40, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=298; total time= 0.5s [CV] END bootstrap=False, max_depth=40, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=298; total time= 0.5s [CV] END bootstrap=False, max_depth=40, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=298; total time= 0.5s [CV] END bootstrap=False, max_depth=40, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=298; total time= 0.5s [CV] END bootstrap=False, max_depth=40, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=298; total time= 0.5s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=91; total time= 0.1s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=91; total time= 0.1s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=91; total time= 0.1s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=91; total time= 0.1s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=91; total time= 0.1s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=472; total time= 0.7s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=472; total time= 0.7s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=472; total time= 0.7s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=472; total time= 0.7s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=472; total time= 0.7s [CV] END bootstrap=False, max_depth=235, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=715; total time= 1.0s [CV] END bootstrap=False, max_depth=235, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=715; total time= 1.1s [CV] END bootstrap=False, max_depth=235, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=715; total time= 1.1s [CV] END bootstrap=False, max_depth=235, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=715; total time= 1.0s [CV] END bootstrap=False, max_depth=235, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=715; total time= 1.0s [CV] END bootstrap=True, max_depth=180, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=328; total time= 0.5s [CV] END bootstrap=True, max_depth=180, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=328; total time= 0.5s [CV] END bootstrap=True, max_depth=180, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=328; total time= 0.6s [CV] END bootstrap=True, max_depth=180, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=328; total time= 0.5s [CV] END bootstrap=True, max_depth=180, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=328; total time= 0.5s [CV] END bootstrap=True, max_depth=15, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=187; total time= 0.2s [CV] END bootstrap=True, max_depth=15, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=187; total time= 0.3s [CV] END bootstrap=True, max_depth=15, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=187; total time= 0.2s [CV] END bootstrap=True, max_depth=15, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=187; total time= 0.2s [CV] END bootstrap=True, max_depth=15, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=187; total time= 0.2s [CV] END bootstrap=True, max_depth=230, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=982; total time= 1.2s [CV] END bootstrap=True, max_depth=230, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=982; total time= 1.2s [CV] END bootstrap=True, max_depth=230, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=982; total time= 1.2s [CV] END bootstrap=True, max_depth=230, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=982; total time= 1.2s [CV] END bootstrap=True, max_depth=230, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=982; total time= 1.2s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=979; total time= 1.2s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=979; total time= 1.3s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=979; total time= 1.3s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=979; total time= 1.2s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=979; total time= 1.2s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=478; total time= 0.7s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=478; total time= 0.7s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=478; total time= 0.7s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=478; total time= 0.7s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=478; total time= 0.7s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=466; total time= 0.7s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=466; total time= 0.8s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=466; total time= 0.6s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=466; total time= 0.8s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=466; total time= 0.7s [CV] END bootstrap=True, max_depth=140, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=685; total time= 0.9s [CV] END bootstrap=True, max_depth=140, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=685; total time= 0.9s [CV] END bootstrap=True, max_depth=140, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=685; total time= 1.0s [CV] END bootstrap=True, max_depth=140, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=685; total time= 0.9s [CV] END bootstrap=True, max_depth=140, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=685; total time= 1.0s [CV] END bootstrap=False, max_depth=85, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=394; total time= 0.5s [CV] END bootstrap=False, max_depth=85, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=394; total time= 0.4s [CV] END bootstrap=False, max_depth=85, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=394; total time= 0.5s [CV] END bootstrap=False, max_depth=85, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=394; total time= 0.5s [CV] END bootstrap=False, max_depth=85, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=394; total time= 0.4s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=253; total time= 0.3s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=253; total time= 0.3s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=253; total time= 0.3s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=253; total time= 0.3s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=253; total time= 0.3s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=736; total time= 1.0s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=736; total time= 1.0s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=736; total time= 1.0s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=736; total time= 0.9s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=736; total time= 0.9s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=781; total time= 0.9s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=781; total time= 0.9s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=781; total time= 0.9s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=781; total time= 0.8s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=781; total time= 0.9s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=736; total time= 1.0s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=736; total time= 1.0s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=736; total time= 0.9s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=736; total time= 1.0s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=736; total time= 0.9s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=301; total time= 0.4s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=301; total time= 0.5s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=301; total time= 0.4s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=301; total time= 0.4s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=301; total time= 0.4s [CV] END bootstrap=False, max_depth=230, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=796; total time= 0.9s [CV] END bootstrap=False, max_depth=230, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=796; total time= 0.9s [CV] END bootstrap=False, max_depth=230, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=796; total time= 1.0s [CV] END bootstrap=False, max_depth=230, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=796; total time= 0.9s [CV] END bootstrap=False, max_depth=230, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=796; total time= 1.0s [CV] END bootstrap=False, max_depth=150, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=220; total time= 0.3s [CV] END bootstrap=False, max_depth=150, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=220; total time= 0.3s [CV] END bootstrap=False, max_depth=150, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=220; total time= 0.3s [CV] END bootstrap=False, max_depth=150, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=220; total time= 0.3s [CV] END bootstrap=False, max_depth=150, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=220; total time= 0.4s [CV] END bootstrap=False, max_depth=65, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=439; total time= 0.5s [CV] END bootstrap=False, max_depth=65, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=439; total time= 0.6s [CV] END bootstrap=False, max_depth=65, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=439; total time= 0.6s [CV] END bootstrap=False, max_depth=65, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=439; total time= 0.6s [CV] END bootstrap=False, max_depth=65, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=439; total time= 0.7s [CV] END bootstrap=True, max_depth=190, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=649; total time= 0.9s [CV] END bootstrap=True, max_depth=190, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=649; total time= 1.0s [CV] END bootstrap=True, max_depth=190, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=649; total time= 1.0s [CV] END bootstrap=True, max_depth=190, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=649; total time= 1.0s [CV] END bootstrap=True, max_depth=190, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=649; total time= 1.1s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=313; total time= 0.4s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=313; total time= 0.5s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=313; total time= 0.4s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=313; total time= 0.5s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=313; total time= 0.5s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=403; total time= 0.5s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=403; total time= 0.5s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=403; total time= 0.4s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=403; total time= 0.5s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=403; total time= 0.4s [CV] END bootstrap=False, max_depth=20, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=598; total time= 0.9s [CV] END bootstrap=False, max_depth=20, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=598; total time= 0.7s [CV] END bootstrap=False, max_depth=20, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=598; total time= 0.9s [CV] END bootstrap=False, max_depth=20, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=598; total time= 0.8s [CV] END bootstrap=False, max_depth=20, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=598; total time= 0.8s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=121; total time= 0.1s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=121; total time= 0.2s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=121; total time= 0.2s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=121; total time= 0.2s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=121; total time= 0.1s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=877; total time= 1.2s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=877; total time= 1.2s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=877; total time= 1.1s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=877; total time= 1.1s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=877; total time= 1.2s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=463; total time= 0.7s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=463; total time= 0.7s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=463; total time= 0.7s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=463; total time= 0.7s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=463; total time= 0.6s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=379; total time= 0.4s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=379; total time= 0.4s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=379; total time= 0.4s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=379; total time= 0.3s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=379; total time= 0.4s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=658; total time= 0.8s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=658; total time= 0.8s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=658; total time= 0.8s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=658; total time= 0.8s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=658; total time= 0.9s [CV] END bootstrap=False, max_depth=55, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=742; total time= 0.9s [CV] END bootstrap=False, max_depth=55, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=742; total time= 1.0s [CV] END bootstrap=False, max_depth=55, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=742; total time= 1.0s [CV] END bootstrap=False, max_depth=55, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=742; total time= 0.9s [CV] END bootstrap=False, max_depth=55, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=742; total time= 0.9s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=403; total time= 0.4s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=403; total time= 0.6s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=403; total time= 0.6s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=403; total time= 0.5s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=403; total time= 0.5s [CV] END bootstrap=False, max_depth=215, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=25; total time= 0.0s [CV] END bootstrap=False, max_depth=215, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=25; total time= 0.0s [CV] END bootstrap=False, max_depth=215, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=25; total time= 0.0s [CV] END bootstrap=False, max_depth=215, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=25; total time= 0.0s [CV] END bootstrap=False, max_depth=215, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=25; total time= 0.0s [CV] END bootstrap=True, max_depth=15, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=544; total time= 0.7s [CV] END bootstrap=True, max_depth=15, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=544; total time= 0.7s [CV] END bootstrap=True, max_depth=15, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=544; total time= 0.7s [CV] END bootstrap=True, max_depth=15, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=544; total time= 0.7s [CV] END bootstrap=True, max_depth=15, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=544; total time= 0.7s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=460; total time= 0.7s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=460; total time= 0.7s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=460; total time= 0.6s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=460; total time= 0.7s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=460; total time= 0.7s [CV] END bootstrap=False, max_depth=95, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=442; total time= 0.7s [CV] END bootstrap=False, max_depth=95, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=442; total time= 0.7s [CV] END bootstrap=False, max_depth=95, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=442; total time= 0.6s [CV] END bootstrap=False, max_depth=95, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=442; total time= 0.6s [CV] END bootstrap=False, max_depth=95, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=442; total time= 0.7s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=676; total time= 0.9s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=676; total time= 0.8s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=676; total time= 0.8s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=676; total time= 0.8s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=676; total time= 1.0s [CV] END bootstrap=False, max_depth=145, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=697; total time= 1.0s [CV] END bootstrap=False, max_depth=145, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=697; total time= 0.9s [CV] END bootstrap=False, max_depth=145, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=697; total time= 0.8s [CV] END bootstrap=False, max_depth=145, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=697; total time= 1.2s [CV] END bootstrap=False, max_depth=145, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=697; total time= 0.9s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=772; total time= 0.9s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=772; total time= 0.9s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=772; total time= 0.8s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=772; total time= 0.9s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=772; total time= 0.9s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=325; total time= 0.5s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=325; total time= 0.5s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=325; total time= 0.4s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=325; total time= 0.5s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=325; total time= 0.5s [CV] END bootstrap=True, max_depth=160, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=667; total time= 0.9s [CV] END bootstrap=True, max_depth=160, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=667; total time= 0.8s [CV] END bootstrap=True, max_depth=160, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=667; total time= 0.8s [CV] END bootstrap=True, max_depth=160, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=667; total time= 0.9s [CV] END bootstrap=True, max_depth=160, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=667; total time= 0.9s [CV] END bootstrap=True, max_depth=245, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=691; total time= 0.9s [CV] END bootstrap=True, max_depth=245, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=691; total time= 0.9s [CV] END bootstrap=True, max_depth=245, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=691; total time= 0.9s [CV] END bootstrap=True, max_depth=245, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=691; total time= 0.9s [CV] END bootstrap=True, max_depth=245, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=691; total time= 0.9s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=805; total time= 0.9s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=805; total time= 0.9s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=805; total time= 0.9s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=805; total time= 0.7s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=805; total time= 0.9s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=211; total time= 0.3s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=211; total time= 0.3s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=211; total time= 0.4s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=211; total time= 0.3s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=211; total time= 0.3s [CV] END bootstrap=False, max_depth=25, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=493; total time= 0.8s [CV] END bootstrap=False, max_depth=25, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=493; total time= 0.9s [CV] END bootstrap=False, max_depth=25, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=493; total time= 0.8s [CV] END bootstrap=False, max_depth=25, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=493; total time= 0.8s [CV] END bootstrap=False, max_depth=25, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=493; total time= 0.8s [CV] END bootstrap=False, max_depth=220, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=307; total time= 0.5s [CV] END bootstrap=False, max_depth=220, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=307; total time= 0.5s [CV] END bootstrap=False, max_depth=220, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=307; total time= 0.5s [CV] END bootstrap=False, max_depth=220, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=307; total time= 0.5s [CV] END bootstrap=False, max_depth=220, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=307; total time= 0.4s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=331; total time= 0.6s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=331; total time= 0.6s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=331; total time= 0.6s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=331; total time= 0.5s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=331; total time= 0.5s [CV] END bootstrap=False, max_depth=130, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=865; total time= 1.0s [CV] END bootstrap=False, max_depth=130, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=865; total time= 1.2s [CV] END bootstrap=False, max_depth=130, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=865; total time= 1.0s [CV] END bootstrap=False, max_depth=130, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=865; total time= 1.0s [CV] END bootstrap=False, max_depth=130, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=865; total time= 1.1s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=166; total time= 0.3s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=166; total time= 0.3s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=166; total time= 0.3s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=166; total time= 0.3s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=166; total time= 0.3s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=985; total time= 1.2s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=985; total time= 1.2s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=985; total time= 1.1s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=985; total time= 1.3s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=985; total time= 1.3s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=694; total time= 0.9s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=694; total time= 0.9s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=694; total time= 0.9s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=694; total time= 0.9s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=694; total time= 0.9s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=118; total time= 0.1s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=118; total time= 0.1s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=118; total time= 0.2s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=118; total time= 0.2s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=118; total time= 0.1s [CV] END bootstrap=False, max_depth=90, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=484; total time= 0.8s [CV] END bootstrap=False, max_depth=90, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=484; total time= 0.7s [CV] END bootstrap=False, max_depth=90, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=484; total time= 0.7s [CV] END bootstrap=False, max_depth=90, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=484; total time= 0.8s [CV] END bootstrap=False, max_depth=90, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=484; total time= 0.8s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=232; total time= 0.4s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=232; total time= 0.4s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=232; total time= 0.4s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=232; total time= 0.3s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=232; total time= 0.3s [CV] END bootstrap=True, max_depth=100, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=343; total time= 0.6s [CV] END bootstrap=True, max_depth=100, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=343; total time= 0.6s [CV] END bootstrap=True, max_depth=100, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=343; total time= 0.6s [CV] END bootstrap=True, max_depth=100, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=343; total time= 0.6s [CV] END bootstrap=True, max_depth=100, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=343; total time= 0.6s [CV] END bootstrap=False, max_depth=255, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=241; total time= 0.3s [CV] END bootstrap=False, max_depth=255, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=241; total time= 0.3s [CV] END bootstrap=False, max_depth=255, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=241; total time= 0.3s [CV] END bootstrap=False, max_depth=255, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=241; total time= 0.3s [CV] END bootstrap=False, max_depth=255, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=241; total time= 0.3s [CV] END bootstrap=False, max_depth=90, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=229; total time= 0.4s [CV] END bootstrap=False, max_depth=90, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=229; total time= 0.3s [CV] END bootstrap=False, max_depth=90, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=229; total time= 0.3s [CV] END bootstrap=False, max_depth=90, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=229; total time= 0.3s [CV] END bootstrap=False, max_depth=90, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=229; total time= 0.4s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=463; total time= 0.8s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=463; total time= 0.7s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=463; total time= 0.8s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=463; total time= 0.7s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=463; total time= 0.7s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=625; total time= 0.9s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=625; total time= 0.9s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=625; total time= 1.0s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=625; total time= 1.0s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=625; total time= 1.0s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=874; total time= 1.2s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=874; total time= 1.0s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=874; total time= 1.2s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=874; total time= 1.2s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=874; total time= 1.1s [CV] END bootstrap=True, max_depth=230, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=61; total time= 0.1s [CV] END bootstrap=True, max_depth=230, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=61; total time= 0.1s [CV] END bootstrap=True, max_depth=230, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=61; total time= 0.1s [CV] END bootstrap=True, max_depth=230, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=61; total time= 0.1s [CV] END bootstrap=True, max_depth=230, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=61; total time= 0.1s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=223; total time= 0.3s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=223; total time= 0.3s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=223; total time= 0.4s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=223; total time= 0.3s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=223; total time= 0.6s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=616; total time= 0.9s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=616; total time= 0.9s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=616; total time= 0.8s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=616; total time= 0.8s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=616; total time= 0.8s [CV] END bootstrap=False, max_depth=250, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=289; total time= 0.4s [CV] END bootstrap=False, max_depth=250, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=289; total time= 0.5s [CV] END bootstrap=False, max_depth=250, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=289; total time= 0.5s [CV] END bootstrap=False, max_depth=250, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=289; total time= 0.5s [CV] END bootstrap=False, max_depth=250, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=289; total time= 0.3s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=241; total time= 0.3s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=241; total time= 0.4s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=241; total time= 0.3s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=241; total time= 0.4s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=241; total time= 0.3s [CV] END bootstrap=False, max_depth=90, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=535; total time= 0.7s [CV] END bootstrap=False, max_depth=90, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=535; total time= 0.7s [CV] END bootstrap=False, max_depth=90, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=535; total time= 0.6s [CV] END bootstrap=False, max_depth=90, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=535; total time= 0.6s [CV] END bootstrap=False, max_depth=90, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=535; total time= 0.6s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=535; total time= 0.6s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=535; total time= 0.6s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=535; total time= 0.6s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=535; total time= 0.6s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=535; total time= 0.6s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=643; total time= 1.0s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=643; total time= 1.0s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=643; total time= 0.9s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=643; total time= 1.0s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=643; total time= 1.0s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=400; total time= 0.5s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=400; total time= 0.4s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=400; total time= 0.4s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=400; total time= 0.4s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=400; total time= 0.5s [CV] END bootstrap=False, max_depth=220, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=256; total time= 0.2s [CV] END bootstrap=False, max_depth=220, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=256; total time= 0.4s [CV] END bootstrap=False, max_depth=220, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=256; total time= 0.3s [CV] END bootstrap=False, max_depth=220, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=256; total time= 0.4s [CV] END bootstrap=False, max_depth=220, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=256; total time= 0.3s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=514; total time= 0.5s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=514; total time= 0.5s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=514; total time= 0.6s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=514; total time= 0.5s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=514; total time= 0.5s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=415; total time= 0.5s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=415; total time= 0.5s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=415; total time= 0.5s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=415; total time= 0.5s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=415; total time= 0.5s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=352; total time= 0.6s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=352; total time= 0.7s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=352; total time= 0.6s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=352; total time= 0.7s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=352; total time= 0.6s [CV] END bootstrap=False, max_depth=230, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=349; total time= 0.7s [CV] END bootstrap=False, max_depth=230, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=349; total time= 0.6s [CV] END bootstrap=False, max_depth=230, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=349; total time= 0.6s [CV] END bootstrap=False, max_depth=230, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=349; total time= 0.7s [CV] END bootstrap=False, max_depth=230, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=349; total time= 0.6s [CV] END bootstrap=False, max_depth=55, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=187; total time= 0.3s [CV] END bootstrap=False, max_depth=55, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=187; total time= 0.2s [CV] END bootstrap=False, max_depth=55, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=187; total time= 0.2s [CV] END bootstrap=False, max_depth=55, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=187; total time= 0.2s [CV] END bootstrap=False, max_depth=55, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=187; total time= 0.2s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=418; total time= 0.6s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=418; total time= 0.5s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=418; total time= 0.5s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=418; total time= 0.6s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=418; total time= 0.6s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=31; total time= 0.0s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=31; total time= 0.0s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=31; total time= 0.0s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=31; total time= 0.0s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=31; total time= 0.0s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=481; total time= 0.7s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=481; total time= 0.7s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=481; total time= 0.7s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=481; total time= 0.7s [CV] END bootstrap=False, max_depth=240, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=481; total time= 0.8s [CV] END bootstrap=True, max_depth=10, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=988; total time= 1.3s [CV] END bootstrap=True, max_depth=10, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=988; total time= 1.3s [CV] END bootstrap=True, max_depth=10, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=988; total time= 1.2s [CV] END bootstrap=True, max_depth=10, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=988; total time= 1.3s [CV] END bootstrap=True, max_depth=10, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=988; total time= 1.2s [CV] END bootstrap=False, max_depth=225, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=523; total time= 0.6s [CV] END bootstrap=False, max_depth=225, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=523; total time= 0.6s [CV] END bootstrap=False, max_depth=225, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=523; total time= 0.6s [CV] END bootstrap=False, max_depth=225, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=523; total time= 0.6s [CV] END bootstrap=False, max_depth=225, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=523; total time= 0.5s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=391; total time= 0.4s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=391; total time= 0.5s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=391; total time= 0.5s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=391; total time= 0.4s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=391; total time= 0.4s [CV] END bootstrap=False, max_depth=210, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=916; total time= 1.1s [CV] END bootstrap=False, max_depth=210, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=916; total time= 1.2s [CV] END bootstrap=False, max_depth=210, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=916; total time= 1.1s [CV] END bootstrap=False, max_depth=210, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=916; total time= 1.2s [CV] END bootstrap=False, max_depth=210, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=916; total time= 1.1s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=211; total time= 0.3s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=211; total time= 0.3s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=211; total time= 0.3s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=211; total time= 0.3s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=211; total time= 0.3s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=271; total time= 0.4s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=271; total time= 0.4s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=271; total time= 0.4s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=271; total time= 0.3s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=271; total time= 0.4s [CV] END bootstrap=True, max_depth=175, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=190; total time= 0.2s [CV] END bootstrap=True, max_depth=175, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=190; total time= 0.2s [CV] END bootstrap=True, max_depth=175, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=190; total time= 0.2s [CV] END bootstrap=True, max_depth=175, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=190; total time= 0.3s [CV] END bootstrap=True, max_depth=175, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=190; total time= 0.2s [CV] END bootstrap=True, max_depth=35, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=160; total time= 0.3s [CV] END bootstrap=True, max_depth=35, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=160; total time= 0.3s [CV] END bootstrap=True, max_depth=35, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=160; total time= 0.3s [CV] END bootstrap=True, max_depth=35, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=160; total time= 0.3s [CV] END bootstrap=True, max_depth=35, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=160; total time= 0.3s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=940; total time= 1.2s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=940; total time= 1.2s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=940; total time= 1.0s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=940; total time= 1.0s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=940; total time= 1.1s [CV] END bootstrap=False, max_depth=10, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=142; total time= 0.2s [CV] END bootstrap=False, max_depth=10, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=142; total time= 0.2s [CV] END bootstrap=False, max_depth=10, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=142; total time= 0.2s [CV] END bootstrap=False, max_depth=10, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=142; total time= 0.2s [CV] END bootstrap=False, max_depth=10, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=142; total time= 0.2s [CV] END bootstrap=True, max_depth=230, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=841; total time= 1.0s [CV] END bootstrap=True, max_depth=230, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=841; total time= 0.9s [CV] END bootstrap=True, max_depth=230, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=841; total time= 1.1s [CV] END bootstrap=True, max_depth=230, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=841; total time= 1.0s [CV] END bootstrap=True, max_depth=230, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=841; total time= 1.0s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=166; total time= 0.3s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=166; total time= 0.3s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=166; total time= 0.3s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=166; total time= 0.3s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=166; total time= 0.4s [CV] END bootstrap=False, max_depth=70, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=226; total time= 0.4s [CV] END bootstrap=False, max_depth=70, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=226; total time= 0.3s [CV] END bootstrap=False, max_depth=70, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=226; total time= 0.4s [CV] END bootstrap=False, max_depth=70, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=226; total time= 0.3s [CV] END bootstrap=False, max_depth=70, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=226; total time= 0.4s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=325; total time= 0.5s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=325; total time= 0.5s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=325; total time= 0.5s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=325; total time= 0.5s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=325; total time= 0.5s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=316; total time= 0.6s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=316; total time= 0.5s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=316; total time= 0.6s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=316; total time= 0.4s [CV] END bootstrap=False, max_depth=240, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=316; total time= 0.5s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=946; total time= 1.1s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=946; total time= 1.1s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=946; total time= 1.2s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=946; total time= 1.3s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=946; total time= 1.0s [CV] END bootstrap=False, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=238; total time= 0.3s [CV] END bootstrap=False, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=238; total time= 0.3s [CV] END bootstrap=False, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=238; total time= 0.4s [CV] END bootstrap=False, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=238; total time= 0.3s [CV] END bootstrap=False, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=238; total time= 0.4s [CV] END bootstrap=False, max_depth=65, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=181; total time= 0.2s [CV] END bootstrap=False, max_depth=65, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=181; total time= 0.2s [CV] END bootstrap=False, max_depth=65, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=181; total time= 0.2s [CV] END bootstrap=False, max_depth=65, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=181; total time= 0.2s [CV] END bootstrap=False, max_depth=65, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=181; total time= 0.2s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=148; total time= 0.2s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=148; total time= 0.2s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=148; total time= 0.2s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=148; total time= 0.2s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=148; total time= 0.2s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=595; total time= 0.9s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=595; total time= 0.7s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=595; total time= 0.8s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=595; total time= 0.8s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=595; total time= 0.8s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=85; total time= 0.1s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=85; total time= 0.1s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=85; total time= 0.1s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=85; total time= 0.1s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=85; total time= 0.1s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=898; total time= 1.2s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=898; total time= 1.1s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=898; total time= 1.3s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=898; total time= 1.2s [CV] END bootstrap=True, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=898; total time= 0.9s [CV] END bootstrap=False, max_depth=180, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=412; total time= 0.5s [CV] END bootstrap=False, max_depth=180, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=412; total time= 0.6s [CV] END bootstrap=False, max_depth=180, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=412; total time= 0.5s [CV] END bootstrap=False, max_depth=180, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=412; total time= 0.5s [CV] END bootstrap=False, max_depth=180, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=412; total time= 0.5s [CV] END bootstrap=True, max_depth=180, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=568; total time= 0.7s [CV] END bootstrap=True, max_depth=180, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=568; total time= 0.7s [CV] END bootstrap=True, max_depth=180, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=568; total time= 0.6s [CV] END bootstrap=True, max_depth=180, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=568; total time= 0.6s [CV] END bootstrap=True, max_depth=180, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=568; total time= 0.7s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=931; total time= 1.1s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=931; total time= 1.1s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=931; total time= 1.1s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=931; total time= 1.1s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=931; total time= 1.0s [CV] END bootstrap=True, max_depth=130, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=289; total time= 0.4s [CV] END bootstrap=True, max_depth=130, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=289; total time= 0.4s [CV] END bootstrap=True, max_depth=130, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=289; total time= 0.4s [CV] END bootstrap=True, max_depth=130, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=289; total time= 0.5s [CV] END bootstrap=True, max_depth=130, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=289; total time= 0.4s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=871; total time= 1.1s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=871; total time= 1.1s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=871; total time= 1.1s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=871; total time= 1.1s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=871; total time= 1.1s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=745; total time= 1.0s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=745; total time= 1.0s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=745; total time= 0.9s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=745; total time= 0.9s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=745; total time= 1.0s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=532; total time= 0.7s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=532; total time= 0.6s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=532; total time= 0.7s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=532; total time= 0.7s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=532; total time= 0.6s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=214; total time= 0.3s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=214; total time= 0.3s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=214; total time= 0.3s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=214; total time= 0.3s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=214; total time= 0.3s [CV] END bootstrap=True, max_depth=85, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=913; total time= 1.1s [CV] END bootstrap=True, max_depth=85, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=913; total time= 1.1s [CV] END bootstrap=True, max_depth=85, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=913; total time= 1.1s [CV] END bootstrap=True, max_depth=85, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=913; total time= 1.2s [CV] END bootstrap=True, max_depth=85, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=913; total time= 1.1s [CV] END bootstrap=False, max_depth=25, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=298; total time= 0.5s [CV] END bootstrap=False, max_depth=25, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=298; total time= 0.5s [CV] END bootstrap=False, max_depth=25, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=298; total time= 0.4s [CV] END bootstrap=False, max_depth=25, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=298; total time= 0.5s [CV] END bootstrap=False, max_depth=25, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=298; total time= 0.5s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=610; total time= 0.9s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=610; total time= 0.8s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=610; total time= 0.8s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=610; total time= 0.8s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=610; total time= 1.0s [CV] END bootstrap=False, max_depth=55, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=931; total time= 1.1s [CV] END bootstrap=False, max_depth=55, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=931; total time= 1.2s [CV] END bootstrap=False, max_depth=55, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=931; total time= 1.0s [CV] END bootstrap=False, max_depth=55, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=931; total time= 1.0s [CV] END bootstrap=False, max_depth=55, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=931; total time= 1.3s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=559; total time= 0.7s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=559; total time= 0.8s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=559; total time= 0.7s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=559; total time= 0.7s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=559; total time= 0.6s [CV] END bootstrap=True, max_depth=110, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=997; total time= 1.1s [CV] END bootstrap=True, max_depth=110, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=997; total time= 1.1s [CV] END bootstrap=True, max_depth=110, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=997; total time= 1.3s [CV] END bootstrap=True, max_depth=110, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=997; total time= 1.1s [CV] END bootstrap=True, max_depth=110, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=997; total time= 1.2s [CV] END bootstrap=True, max_depth=190, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=82; total time= 0.1s [CV] END bootstrap=True, max_depth=190, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=82; total time= 0.1s [CV] END bootstrap=True, max_depth=190, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=82; total time= 0.1s [CV] END bootstrap=True, max_depth=190, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=82; total time= 0.1s [CV] END bootstrap=True, max_depth=190, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=82; total time= 0.1s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=61; total time= 0.1s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=61; total time= 0.1s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=61; total time= 0.1s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=61; total time= 0.1s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=61; total time= 0.1s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=241; total time= 0.3s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=241; total time= 0.3s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=241; total time= 0.3s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=241; total time= 0.4s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=241; total time= 0.3s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=643; total time= 1.0s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=643; total time= 1.0s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=643; total time= 1.0s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=643; total time= 0.9s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=643; total time= 1.0s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=64; total time= 0.0s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=64; total time= 0.1s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=64; total time= 0.1s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=64; total time= 0.1s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=64; total time= 0.1s [CV] END bootstrap=False, max_depth=110, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=943; total time= 1.1s [CV] END bootstrap=False, max_depth=110, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=943; total time= 1.1s [CV] END bootstrap=False, max_depth=110, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=943; total time= 1.2s [CV] END bootstrap=False, max_depth=110, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=943; total time= 1.3s [CV] END bootstrap=False, max_depth=110, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=943; total time= 1.2s [CV] END bootstrap=False, max_depth=115, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=412; total time= 0.6s [CV] END bootstrap=False, max_depth=115, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=412; total time= 0.6s [CV] END bootstrap=False, max_depth=115, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=412; total time= 0.5s [CV] END bootstrap=False, max_depth=115, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=412; total time= 0.6s [CV] END bootstrap=False, max_depth=115, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=412; total time= 0.4s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=592; total time= 0.8s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=592; total time= 0.8s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=592; total time= 0.8s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=592; total time= 0.9s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=592; total time= 0.9s [CV] END bootstrap=False, max_depth=135, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=346; total time= 0.5s [CV] END bootstrap=False, max_depth=135, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=346; total time= 0.6s [CV] END bootstrap=False, max_depth=135, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=346; total time= 0.6s [CV] END bootstrap=False, max_depth=135, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=346; total time= 0.6s [CV] END bootstrap=False, max_depth=135, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=346; total time= 0.6s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=442; total time= 0.7s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=442; total time= 0.6s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=442; total time= 0.6s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=442; total time= 0.6s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=442; total time= 0.6s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=199; total time= 0.3s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=199; total time= 0.3s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=199; total time= 0.2s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=199; total time= 0.3s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=199; total time= 0.3s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=901; total time= 1.3s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=901; total time= 1.3s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=901; total time= 1.2s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=901; total time= 1.1s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=901; total time= 1.2s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=595; total time= 0.8s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=595; total time= 0.7s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=595; total time= 0.8s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=595; total time= 0.8s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=595; total time= 0.8s [CV] END bootstrap=False, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=286; total time= 0.5s [CV] END bootstrap=False, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=286; total time= 0.4s [CV] END bootstrap=False, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=286; total time= 0.4s [CV] END bootstrap=False, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=286; total time= 0.4s [CV] END bootstrap=False, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=286; total time= 0.4s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=283; total time= 0.4s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=283; total time= 0.4s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=283; total time= 0.4s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=283; total time= 0.4s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=283; total time= 0.3s [CV] END bootstrap=False, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=814; total time= 0.9s [CV] END bootstrap=False, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=814; total time= 1.0s [CV] END bootstrap=False, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=814; total time= 1.1s [CV] END bootstrap=False, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=814; total time= 1.0s [CV] END bootstrap=False, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=814; total time= 1.0s [CV] END bootstrap=False, max_depth=190, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=55; total time= 0.1s [CV] END bootstrap=False, max_depth=190, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=55; total time= 0.1s [CV] END bootstrap=False, max_depth=190, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=55; total time= 0.1s [CV] END bootstrap=False, max_depth=190, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=55; total time= 0.1s [CV] END bootstrap=False, max_depth=190, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=55; total time= 0.1s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=694; total time= 0.9s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=694; total time= 0.9s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=694; total time= 0.9s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=694; total time= 1.0s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=694; total time= 0.8s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=475; total time= 0.7s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=475; total time= 0.8s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=475; total time= 0.8s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=475; total time= 0.7s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=475; total time= 0.8s [CV] END bootstrap=False, max_depth=180, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=31; total time= 0.0s [CV] END bootstrap=False, max_depth=180, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=31; total time= 0.0s [CV] END bootstrap=False, max_depth=180, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=31; total time= 0.0s [CV] END bootstrap=False, max_depth=180, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=31; total time= 0.0s [CV] END bootstrap=False, max_depth=180, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=31; total time= 0.0s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=898; total time= 1.1s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=898; total time= 1.2s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=898; total time= 1.1s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=898; total time= 1.3s [CV] END bootstrap=False, max_depth=250, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=898; total time= 1.2s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=700; total time= 0.9s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=700; total time= 0.9s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=700; total time= 0.8s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=700; total time= 1.1s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=700; total time= 0.9s [CV] END bootstrap=True, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=289; total time= 0.4s [CV] END bootstrap=True, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=289; total time= 0.4s [CV] END bootstrap=True, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=289; total time= 0.4s [CV] END bootstrap=True, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=289; total time= 0.3s [CV] END bootstrap=True, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=289; total time= 0.4s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=931; total time= 1.2s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=931; total time= 1.0s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=931; total time= 1.3s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=931; total time= 1.2s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=931; total time= 1.1s [CV] END bootstrap=False, max_depth=245, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=463; total time= 0.7s [CV] END bootstrap=False, max_depth=245, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=463; total time= 0.6s [CV] END bootstrap=False, max_depth=245, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=463; total time= 0.7s [CV] END bootstrap=False, max_depth=245, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=463; total time= 0.7s [CV] END bootstrap=False, max_depth=245, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=463; total time= 0.7s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=178; total time= 0.3s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=178; total time= 0.2s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=178; total time= 0.2s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=178; total time= 0.2s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=178; total time= 0.3s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=76; total time= 0.2s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=76; total time= 0.1s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=76; total time= 0.1s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=76; total time= 0.1s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=76; total time= 0.2s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=40; total time= 0.0s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=40; total time= 0.0s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=40; total time= 0.0s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=40; total time= 0.0s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=40; total time= 0.0s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=517; total time= 0.5s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=517; total time= 0.6s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=517; total time= 0.6s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=517; total time= 0.6s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=517; total time= 0.5s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=259; total time= 0.3s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=259; total time= 0.4s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=259; total time= 0.3s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=259; total time= 0.4s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=259; total time= 0.4s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=373; total time= 0.4s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=373; total time= 0.5s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=373; total time= 0.4s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=373; total time= 0.4s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=373; total time= 0.4s [CV] END bootstrap=False, max_depth=20, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=130; total time= 0.2s [CV] END bootstrap=False, max_depth=20, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=130; total time= 0.1s [CV] END bootstrap=False, max_depth=20, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=130; total time= 0.2s [CV] END bootstrap=False, max_depth=20, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=130; total time= 0.2s [CV] END bootstrap=False, max_depth=20, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=130; total time= 0.2s [CV] END bootstrap=False, max_depth=100, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=124; total time= 0.1s [CV] END bootstrap=False, max_depth=100, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=124; total time= 0.2s [CV] END bootstrap=False, max_depth=100, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=124; total time= 0.2s [CV] END bootstrap=False, max_depth=100, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=124; total time= 0.2s [CV] END bootstrap=False, max_depth=100, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=124; total time= 0.2s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=697; total time= 1.0s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=697; total time= 1.0s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=697; total time= 1.0s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=697; total time= 1.0s [CV] END bootstrap=True, max_depth=55, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=697; total time= 1.0s [CV] END bootstrap=False, max_depth=160, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=280; total time= 0.4s [CV] END bootstrap=False, max_depth=160, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=280; total time= 0.4s [CV] END bootstrap=False, max_depth=160, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=280; total time= 0.4s [CV] END bootstrap=False, max_depth=160, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=280; total time= 0.4s [CV] END bootstrap=False, max_depth=160, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=280; total time= 0.4s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=466; total time= 0.8s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=466; total time= 0.7s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=466; total time= 0.6s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=466; total time= 0.8s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=466; total time= 0.7s [CV] END bootstrap=True, max_depth=245, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=46; total time= 0.0s [CV] END bootstrap=True, max_depth=245, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=46; total time= 0.1s [CV] END bootstrap=True, max_depth=245, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=46; total time= 0.1s [CV] END bootstrap=True, max_depth=245, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=46; total time= 0.1s [CV] END bootstrap=True, max_depth=245, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=46; total time= 0.1s [CV] END bootstrap=False, max_depth=110, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=328; total time= 0.6s [CV] END bootstrap=False, max_depth=110, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=328; total time= 0.6s [CV] END bootstrap=False, max_depth=110, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=328; total time= 0.5s [CV] END bootstrap=False, max_depth=110, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=328; total time= 0.2s [CV] END bootstrap=False, max_depth=110, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=328; total time= 0.2s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=16; total time= 0.0s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=16; total time= 0.0s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=16; total time= 0.0s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=16; total time= 0.0s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=16; total time= 0.0s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=415; total time= 0.2s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=415; total time= 0.2s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=415; total time= 0.2s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=415; total time= 0.2s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=415; total time= 0.2s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=574; total time= 0.3s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=574; total time= 0.3s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=574; total time= 0.3s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=574; total time= 0.3s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=574; total time= 0.3s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=607; total time= 0.6s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=607; total time= 0.9s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=607; total time= 0.8s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=607; total time= 0.8s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=607; total time= 0.6s [CV] END bootstrap=True, max_depth=40, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=523; total time= 0.5s [CV] END bootstrap=True, max_depth=40, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=523; total time= 0.5s [CV] END bootstrap=True, max_depth=40, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=523; total time= 0.5s [CV] END bootstrap=True, max_depth=40, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=523; total time= 0.5s [CV] END bootstrap=True, max_depth=40, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=523; total time= 0.4s [CV] END bootstrap=False, max_depth=180, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=727; total time= 0.8s [CV] END bootstrap=False, max_depth=180, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=727; total time= 0.7s [CV] END bootstrap=False, max_depth=180, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=727; total time= 0.8s [CV] END bootstrap=False, max_depth=180, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=727; total time= 0.8s [CV] END bootstrap=False, max_depth=180, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=727; total time= 0.8s [CV] END bootstrap=True, max_depth=115, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=328; total time= 0.5s [CV] END bootstrap=True, max_depth=115, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=328; total time= 0.5s [CV] END bootstrap=True, max_depth=115, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=328; total time= 0.5s [CV] END bootstrap=True, max_depth=115, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=328; total time= 0.4s [CV] END bootstrap=True, max_depth=115, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=328; total time= 0.4s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=805; total time= 0.8s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=805; total time= 0.7s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=805; total time= 0.8s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=805; total time= 0.8s [CV] END bootstrap=True, max_depth=85, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=805; total time= 0.8s [CV] END bootstrap=False, max_depth=110, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=967; total time= 0.9s [CV] END bootstrap=False, max_depth=110, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=967; total time= 1.0s [CV] END bootstrap=False, max_depth=110, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=967; total time= 1.0s [CV] END bootstrap=False, max_depth=110, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=967; total time= 0.9s [CV] END bootstrap=False, max_depth=110, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=967; total time= 1.1s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=934; total time= 1.0s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=934; total time= 1.0s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=934; total time= 0.9s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=934; total time= 1.0s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=934; total time= 0.8s [CV] END bootstrap=True, max_depth=75, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=643; total time= 0.8s [CV] END bootstrap=True, max_depth=75, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=643; total time= 0.8s [CV] END bootstrap=True, max_depth=75, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=643; total time= 0.9s [CV] END bootstrap=True, max_depth=75, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=643; total time= 0.9s [CV] END bootstrap=True, max_depth=75, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=643; total time= 0.9s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=835; total time= 0.8s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=835; total time= 0.8s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=835; total time= 0.8s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=835; total time= 0.9s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=835; total time= 0.9s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=169; total time= 0.2s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=169; total time= 0.2s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=169; total time= 0.2s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=169; total time= 0.2s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=169; total time= 0.3s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=391; total time= 0.4s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=391; total time= 0.4s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=391; total time= 0.3s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=391; total time= 0.4s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=391; total time= 0.4s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=442; total time= 0.5s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=442; total time= 0.5s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=442; total time= 0.6s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=442; total time= 0.5s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=442; total time= 0.6s [CV] END bootstrap=True, max_depth=235, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=187; total time= 0.1s [CV] END bootstrap=True, max_depth=235, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=187; total time= 0.2s [CV] END bootstrap=True, max_depth=235, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=187; total time= 0.2s [CV] END bootstrap=True, max_depth=235, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=187; total time= 0.2s [CV] END bootstrap=True, max_depth=235, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=187; total time= 0.2s [CV] END bootstrap=True, max_depth=100, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=895; total time= 1.1s [CV] END bootstrap=True, max_depth=100, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=895; total time= 1.0s [CV] END bootstrap=True, max_depth=100, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=895; total time= 0.9s [CV] END bootstrap=True, max_depth=100, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=895; total time= 1.0s [CV] END bootstrap=True, max_depth=100, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=895; total time= 1.1s [CV] END bootstrap=True, max_depth=245, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=805; total time= 0.6s [CV] END bootstrap=True, max_depth=245, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=805; total time= 0.7s [CV] END bootstrap=True, max_depth=245, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=805; total time= 0.7s [CV] END bootstrap=True, max_depth=245, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=805; total time= 0.8s [CV] END bootstrap=True, max_depth=245, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=805; total time= 0.8s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=1000; total time= 1.0s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=1000; total time= 1.0s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=1000; total time= 0.9s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=1000; total time= 1.1s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=1000; total time= 0.9s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=715; total time= 0.7s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=715; total time= 0.8s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=715; total time= 0.7s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=715; total time= 0.8s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=715; total time= 0.8s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=196; total time= 0.2s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=196; total time= 0.2s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=196; total time= 0.2s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=196; total time= 0.2s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=196; total time= 0.3s [CV] END bootstrap=False, max_depth=235, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=772; total time= 0.6s [CV] END bootstrap=False, max_depth=235, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=772; total time= 0.7s [CV] END bootstrap=False, max_depth=235, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=772; total time= 0.6s [CV] END bootstrap=False, max_depth=235, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=772; total time= 0.7s [CV] END bootstrap=False, max_depth=235, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=772; total time= 0.6s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=889; total time= 1.0s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=889; total time= 0.9s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=889; total time= 1.0s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=889; total time= 1.0s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=889; total time= 1.0s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=586; total time= 0.6s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=586; total time= 0.6s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=586; total time= 0.7s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=586; total time= 0.7s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=586; total time= 0.7s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=829; total time= 0.8s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=829; total time= 0.8s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=829; total time= 0.8s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=829; total time= 0.8s [CV] END bootstrap=False, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=829; total time= 0.8s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=16; total time= 0.0s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=16; total time= 0.0s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=16; total time= 0.0s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=16; total time= 0.0s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=16; total time= 0.0s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=607; total time= 0.8s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=607; total time= 0.7s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=607; total time= 0.7s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=607; total time= 0.7s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=607; total time= 0.7s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=631; total time= 0.8s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=631; total time= 0.8s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=631; total time= 0.8s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=631; total time= 0.8s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=631; total time= 0.9s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=583; total time= 0.7s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=583; total time= 0.6s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=583; total time= 0.6s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=583; total time= 0.6s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=583; total time= 0.7s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=295; total time= 0.3s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=295; total time= 0.3s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=295; total time= 0.3s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=295; total time= 0.3s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=295; total time= 0.4s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=136; total time= 0.2s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=136; total time= 0.2s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=136; total time= 0.2s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=136; total time= 0.2s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=136; total time= 0.2s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=64; total time= 0.1s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=64; total time= 0.1s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=64; total time= 0.1s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=64; total time= 0.1s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=64; total time= 0.1s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=931; total time= 1.0s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=931; total time= 1.0s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=931; total time= 1.0s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=931; total time= 0.9s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=931; total time= 0.8s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=574; total time= 0.6s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=574; total time= 0.6s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=574; total time= 0.6s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=574; total time= 0.7s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=574; total time= 0.6s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=907; total time= 1.2s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=907; total time= 1.0s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=907; total time= 1.0s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=907; total time= 1.2s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=907; total time= 1.1s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=829; total time= 0.9s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=829; total time= 0.9s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=829; total time= 0.8s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=829; total time= 0.8s [CV] END bootstrap=False, max_depth=35, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=829; total time= 0.8s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=703; total time= 0.8s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=703; total time= 0.7s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=703; total time= 0.9s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=703; total time= 0.9s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=703; total time= 0.7s [CV] END bootstrap=True, max_depth=150, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=826; total time= 0.8s [CV] END bootstrap=True, max_depth=150, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=826; total time= 0.8s [CV] END bootstrap=True, max_depth=150, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=826; total time= 0.8s [CV] END bootstrap=True, max_depth=150, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=826; total time= 0.8s [CV] END bootstrap=True, max_depth=150, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=826; total time= 0.8s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=253; total time= 0.2s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=253; total time= 0.2s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=253; total time= 0.2s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=253; total time= 0.2s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=253; total time= 0.2s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=622; total time= 0.8s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=622; total time= 0.8s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=622; total time= 0.8s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=622; total time= 0.8s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=622; total time= 0.8s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=16; total time= 0.0s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=16; total time= 0.0s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=16; total time= 0.0s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=16; total time= 0.0s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=16; total time= 0.0s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=817; total time= 0.8s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=817; total time= 0.7s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=817; total time= 0.8s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=817; total time= 0.9s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=817; total time= 0.7s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=214; total time= 0.3s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=214; total time= 0.2s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=214; total time= 0.2s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=214; total time= 0.3s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=214; total time= 0.2s [CV] END bootstrap=False, max_depth=75, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=202; total time= 0.2s [CV] END bootstrap=False, max_depth=75, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=202; total time= 0.3s [CV] END bootstrap=False, max_depth=75, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=202; total time= 0.2s [CV] END bootstrap=False, max_depth=75, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=202; total time= 0.3s [CV] END bootstrap=False, max_depth=75, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=202; total time= 0.2s [CV] END bootstrap=False, max_depth=65, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=334; total time= 0.4s [CV] END bootstrap=False, max_depth=65, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=334; total time= 0.5s [CV] END bootstrap=False, max_depth=65, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=334; total time= 0.4s [CV] END bootstrap=False, max_depth=65, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=334; total time= 0.4s [CV] END bootstrap=False, max_depth=65, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=334; total time= 0.4s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=559; total time= 0.5s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=559; total time= 0.6s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=559; total time= 0.6s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=559; total time= 0.5s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=559; total time= 0.5s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=928; total time= 0.9s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=928; total time= 1.0s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=928; total time= 0.9s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=928; total time= 0.9s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=928; total time= 0.9s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=553; total time= 0.6s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=553; total time= 0.5s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=553; total time= 0.7s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=553; total time= 0.6s [CV] END bootstrap=False, max_depth=150, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=553; total time= 0.5s [CV] END bootstrap=False, max_depth=115, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=757; total time= 0.7s [CV] END bootstrap=False, max_depth=115, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=757; total time= 0.6s [CV] END bootstrap=False, max_depth=115, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=757; total time= 0.6s [CV] END bootstrap=False, max_depth=115, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=757; total time= 0.6s [CV] END bootstrap=False, max_depth=115, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=757; total time= 0.8s [CV] END bootstrap=False, max_depth=170, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=199; total time= 0.2s [CV] END bootstrap=False, max_depth=170, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=199; total time= 0.3s [CV] END bootstrap=False, max_depth=170, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=199; total time= 0.2s [CV] END bootstrap=False, max_depth=170, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=199; total time= 0.2s [CV] END bootstrap=False, max_depth=170, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=199; total time= 0.3s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=265; total time= 0.3s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=265; total time= 0.3s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=265; total time= 0.2s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=265; total time= 0.3s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=265; total time= 0.3s [CV] END bootstrap=True, max_depth=50, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=964; total time= 1.0s [CV] END bootstrap=True, max_depth=50, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=964; total time= 1.0s [CV] END bootstrap=True, max_depth=50, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=964; total time= 1.0s [CV] END bootstrap=True, max_depth=50, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=964; total time= 1.1s [CV] END bootstrap=True, max_depth=50, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=964; total time= 1.0s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=910; total time= 1.0s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=910; total time= 1.1s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=910; total time= 1.0s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=910; total time= 1.1s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=910; total time= 1.1s [CV] END bootstrap=True, max_depth=165, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=595; total time= 0.8s [CV] END bootstrap=True, max_depth=165, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=595; total time= 0.7s [CV] END bootstrap=True, max_depth=165, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=595; total time= 0.7s [CV] END bootstrap=True, max_depth=165, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=595; total time= 0.7s [CV] END bootstrap=True, max_depth=165, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=595; total time= 0.8s [CV] END bootstrap=True, max_depth=200, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=994; total time= 1.0s [CV] END bootstrap=True, max_depth=200, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=994; total time= 0.9s [CV] END bootstrap=True, max_depth=200, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=994; total time= 0.8s [CV] END bootstrap=True, max_depth=200, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=994; total time= 0.9s [CV] END bootstrap=True, max_depth=200, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=994; total time= 0.9s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=178; total time= 0.3s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=178; total time= 0.3s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=178; total time= 0.2s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=178; total time= 0.2s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=178; total time= 0.1s [CV] END bootstrap=False, max_depth=250, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=838; total time= 0.9s [CV] END bootstrap=False, max_depth=250, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=838; total time= 0.9s [CV] END bootstrap=False, max_depth=250, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=838; total time= 0.9s [CV] END bootstrap=False, max_depth=250, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=838; total time= 0.8s [CV] END bootstrap=False, max_depth=250, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=838; total time= 0.8s [CV] END bootstrap=True, max_depth=230, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=736; total time= 1.0s [CV] END bootstrap=True, max_depth=230, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=736; total time= 0.7s [CV] END bootstrap=True, max_depth=230, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=736; total time= 0.7s [CV] END bootstrap=True, max_depth=230, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=736; total time= 0.8s [CV] END bootstrap=True, max_depth=230, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=736; total time= 0.9s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=562; total time= 0.7s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=562; total time= 0.6s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=562; total time= 0.6s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=562; total time= 0.6s [CV] END bootstrap=True, max_depth=145, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=562; total time= 0.5s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=742; total time= 0.7s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=742; total time= 0.8s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=742; total time= 0.9s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=742; total time= 0.8s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=742; total time= 0.7s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=478; total time= 0.6s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=478; total time= 0.6s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=478; total time= 0.6s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=478; total time= 0.6s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=478; total time= 0.5s [CV] END bootstrap=False, max_depth=70, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=982; total time= 1.0s [CV] END bootstrap=False, max_depth=70, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=982; total time= 0.9s [CV] END bootstrap=False, max_depth=70, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=982; total time= 1.1s [CV] END bootstrap=False, max_depth=70, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=982; total time= 0.9s [CV] END bootstrap=False, max_depth=70, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=982; total time= 0.9s [CV] END bootstrap=False, max_depth=70, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=469; total time= 0.6s [CV] END bootstrap=False, max_depth=70, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=469; total time= 0.6s [CV] END bootstrap=False, max_depth=70, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=469; total time= 0.6s [CV] END bootstrap=False, max_depth=70, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=469; total time= 0.6s [CV] END bootstrap=False, max_depth=70, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=469; total time= 0.6s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=937; total time= 1.0s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=937; total time= 1.0s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=937; total time= 0.9s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=937; total time= 1.0s [CV] END bootstrap=False, max_depth=165, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=937; total time= 1.0s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=886; total time= 1.0s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=886; total time= 1.1s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=886; total time= 0.9s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=886; total time= 0.9s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=886; total time= 1.0s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=484; total time= 0.6s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=484; total time= 0.6s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=484; total time= 0.6s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=484; total time= 0.7s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=484; total time= 0.6s [CV] END bootstrap=False, max_depth=145, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=691; total time= 0.6s [CV] END bootstrap=False, max_depth=145, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=691; total time= 0.8s [CV] END bootstrap=False, max_depth=145, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=691; total time= 0.8s [CV] END bootstrap=False, max_depth=145, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=691; total time= 0.7s [CV] END bootstrap=False, max_depth=145, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=691; total time= 0.8s [CV] END bootstrap=True, max_depth=175, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=445; total time= 0.5s [CV] END bootstrap=True, max_depth=175, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=445; total time= 0.5s [CV] END bootstrap=True, max_depth=175, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=445; total time= 0.5s [CV] END bootstrap=True, max_depth=175, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=445; total time= 0.4s [CV] END bootstrap=True, max_depth=175, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=445; total time= 0.5s [CV] END bootstrap=False, max_depth=210, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=805; total time= 0.9s [CV] END bootstrap=False, max_depth=210, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=805; total time= 0.8s [CV] END bootstrap=False, max_depth=210, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=805; total time= 0.8s [CV] END bootstrap=False, max_depth=210, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=805; total time= 0.8s [CV] END bootstrap=False, max_depth=210, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=805; total time= 0.8s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=421; total time= 0.5s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=421; total time= 0.5s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=421; total time= 0.5s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=421; total time= 0.6s [CV] END bootstrap=False, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=421; total time= 0.5s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=316; total time= 0.5s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=316; total time= 0.4s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=316; total time= 0.4s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=316; total time= 0.4s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=316; total time= 0.4s [CV] END bootstrap=True, max_depth=210, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=388; total time= 0.3s [CV] END bootstrap=True, max_depth=210, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=388; total time= 0.4s [CV] END bootstrap=True, max_depth=210, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=388; total time= 0.3s [CV] END bootstrap=True, max_depth=210, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=388; total time= 0.4s [CV] END bootstrap=True, max_depth=210, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=388; total time= 0.3s [CV] END bootstrap=False, max_depth=40, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=274; total time= 0.2s [CV] END bootstrap=False, max_depth=40, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=274; total time= 0.3s [CV] END bootstrap=False, max_depth=40, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=274; total time= 0.3s [CV] END bootstrap=False, max_depth=40, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=274; total time= 0.3s [CV] END bootstrap=False, max_depth=40, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=274; total time= 0.3s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=370; total time= 0.5s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=370; total time= 0.3s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=370; total time= 0.5s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=370; total time= 0.6s [CV] END bootstrap=True, max_depth=115, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=370; total time= 0.3s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=862; total time= 0.9s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=862; total time= 0.9s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=862; total time= 0.8s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=862; total time= 0.9s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=862; total time= 1.0s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=67; total time= 0.0s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=67; total time= 0.0s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=67; total time= 0.1s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=67; total time= 0.1s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=67; total time= 0.1s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=466; total time= 0.6s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=466; total time= 0.6s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=466; total time= 0.5s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=466; total time= 0.6s [CV] END bootstrap=False, max_depth=120, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=466; total time= 0.6s [CV] END bootstrap=True, max_depth=95, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=181; total time= 0.2s [CV] END bootstrap=True, max_depth=95, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=181; total time= 0.1s [CV] END bootstrap=True, max_depth=95, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=181; total time= 0.2s [CV] END bootstrap=True, max_depth=95, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=181; total time= 0.2s [CV] END bootstrap=True, max_depth=95, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=181; total time= 0.3s [CV] END bootstrap=False, max_depth=90, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=979; total time= 1.1s [CV] END bootstrap=False, max_depth=90, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=979; total time= 1.0s [CV] END bootstrap=False, max_depth=90, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=979; total time= 1.1s [CV] END bootstrap=False, max_depth=90, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=979; total time= 1.0s [CV] END bootstrap=False, max_depth=90, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=979; total time= 0.9s [CV] END bootstrap=True, max_depth=220, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=61; total time= 0.1s [CV] END bootstrap=True, max_depth=220, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=61; total time= 0.0s [CV] END bootstrap=True, max_depth=220, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=61; total time= 0.0s [CV] END bootstrap=True, max_depth=220, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=61; total time= 0.0s [CV] END bootstrap=True, max_depth=220, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=61; total time= 0.1s [CV] END bootstrap=False, max_depth=115, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=208; total time= 0.2s [CV] END bootstrap=False, max_depth=115, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=208; total time= 0.3s [CV] END bootstrap=False, max_depth=115, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=208; total time= 0.3s [CV] END bootstrap=False, max_depth=115, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=208; total time= 0.3s [CV] END bootstrap=False, max_depth=115, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=208; total time= 0.2s [CV] END bootstrap=True, max_depth=110, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=334; total time= 0.4s [CV] END bootstrap=True, max_depth=110, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=334; total time= 0.4s [CV] END bootstrap=True, max_depth=110, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=334; total time= 0.5s [CV] END bootstrap=True, max_depth=110, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=334; total time= 0.4s [CV] END bootstrap=True, max_depth=110, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=334; total time= 0.4s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=667; total time= 0.7s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=667; total time= 0.7s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=667; total time= 0.8s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=667; total time= 0.6s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=667; total time= 0.8s [CV] END bootstrap=True, max_depth=10, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=394; total time= 0.3s [CV] END bootstrap=True, max_depth=10, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=394; total time= 0.4s [CV] END bootstrap=True, max_depth=10, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=394; total time= 0.4s [CV] END bootstrap=True, max_depth=10, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=394; total time= 0.3s [CV] END bootstrap=True, max_depth=10, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=394; total time= 0.4s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=898; total time= 1.0s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=898; total time= 1.0s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=898; total time= 1.0s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=898; total time= 1.0s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=898; total time= 1.0s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=973; total time= 0.9s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=973; total time= 0.9s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=973; total time= 1.0s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=973; total time= 1.0s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=973; total time= 1.1s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=325; total time= 0.4s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=325; total time= 0.5s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=325; total time= 0.6s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=325; total time= 0.5s [CV] END bootstrap=True, max_depth=255, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=325; total time= 0.4s [CV] END bootstrap=True, max_depth=255, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=469; total time= 0.6s [CV] END bootstrap=True, max_depth=255, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=469; total time= 0.6s [CV] END bootstrap=True, max_depth=255, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=469; total time= 0.6s [CV] END bootstrap=True, max_depth=255, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=469; total time= 0.7s [CV] END bootstrap=True, max_depth=255, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=469; total time= 0.6s [CV] END bootstrap=False, max_depth=175, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=955; total time= 0.9s [CV] END bootstrap=False, max_depth=175, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=955; total time= 1.1s [CV] END bootstrap=False, max_depth=175, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=955; total time= 0.9s [CV] END bootstrap=False, max_depth=175, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=955; total time= 1.0s [CV] END bootstrap=False, max_depth=175, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=955; total time= 0.9s [CV] END bootstrap=True, max_depth=10, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=298; total time= 0.3s [CV] END bootstrap=True, max_depth=10, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=298; total time= 0.4s [CV] END bootstrap=True, max_depth=10, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=298; total time= 0.4s [CV] END bootstrap=True, max_depth=10, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=298; total time= 0.3s [CV] END bootstrap=True, max_depth=10, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=298; total time= 0.3s [CV] END bootstrap=True, max_depth=175, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=808; total time= 0.8s [CV] END bootstrap=True, max_depth=175, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=808; total time= 0.7s [CV] END bootstrap=True, max_depth=175, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=808; total time= 0.8s [CV] END bootstrap=True, max_depth=175, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=808; total time= 0.8s [CV] END bootstrap=True, max_depth=175, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=808; total time= 0.6s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=346; total time= 0.4s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=346; total time= 0.5s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=346; total time= 0.4s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=346; total time= 0.6s [CV] END bootstrap=True, max_depth=105, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=346; total time= 0.5s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=976; total time= 1.0s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=976; total time= 1.1s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=976; total time= 1.0s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=976; total time= 1.0s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=976; total time= 1.0s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=403; total time= 0.4s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=403; total time= 0.4s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=403; total time= 0.4s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=403; total time= 0.4s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=403; total time= 0.4s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=628; total time= 0.8s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=628; total time= 0.8s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=628; total time= 0.8s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=628; total time= 0.7s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=628; total time= 0.7s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=826; total time= 0.9s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=826; total time= 0.8s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=826; total time= 0.9s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=826; total time= 0.8s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=826; total time= 0.7s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=343; total time= 0.4s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=343; total time= 0.6s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=343; total time= 0.5s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=343; total time= 0.5s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=343; total time= 0.5s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=433; total time= 0.5s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=433; total time= 0.6s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=433; total time= 0.6s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=433; total time= 0.5s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=433; total time= 0.5s [CV] END bootstrap=True, max_depth=255, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=220; total time= 0.2s [CV] END bootstrap=True, max_depth=255, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=220; total time= 0.3s [CV] END bootstrap=True, max_depth=255, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=220; total time= 0.2s [CV] END bootstrap=True, max_depth=255, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=220; total time= 0.3s [CV] END bootstrap=True, max_depth=255, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=220; total time= 0.3s [CV] END bootstrap=True, max_depth=180, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=244; total time= 0.2s [CV] END bootstrap=True, max_depth=180, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=244; total time= 0.2s [CV] END bootstrap=True, max_depth=180, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=244; total time= 0.2s [CV] END bootstrap=True, max_depth=180, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=244; total time= 0.2s [CV] END bootstrap=True, max_depth=180, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=244; total time= 0.2s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=634; total time= 0.8s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=634; total time= 0.9s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=634; total time= 0.8s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=634; total time= 0.8s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=634; total time= 0.8s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=547; total time= 0.4s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=547; total time= 0.6s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=547; total time= 0.6s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=547; total time= 0.6s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=547; total time= 0.6s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=529; total time= 0.5s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=529; total time= 0.5s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=529; total time= 0.5s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=529; total time= 0.4s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=529; total time= 0.5s [CV] END bootstrap=True, max_depth=150, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=43; total time= 0.1s [CV] END bootstrap=True, max_depth=150, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=43; total time= 0.0s [CV] END bootstrap=True, max_depth=150, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=43; total time= 0.0s [CV] END bootstrap=True, max_depth=150, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=43; total time= 0.0s [CV] END bootstrap=True, max_depth=150, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=43; total time= 0.0s [CV] END bootstrap=True, max_depth=150, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=337; total time= 0.4s [CV] END bootstrap=True, max_depth=150, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=337; total time= 0.5s [CV] END bootstrap=True, max_depth=150, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=337; total time= 0.4s [CV] END bootstrap=True, max_depth=150, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=337; total time= 0.5s [CV] END bootstrap=True, max_depth=150, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=337; total time= 0.5s [CV] END bootstrap=False, max_depth=60, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=700; total time= 0.8s [CV] END bootstrap=False, max_depth=60, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=700; total time= 0.8s [CV] END bootstrap=False, max_depth=60, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=700; total time= 0.8s [CV] END bootstrap=False, max_depth=60, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=700; total time= 0.8s [CV] END bootstrap=False, max_depth=60, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=700; total time= 0.8s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=838; total time= 0.9s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=838; total time= 0.8s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=838; total time= 0.8s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=838; total time= 0.9s [CV] END bootstrap=True, max_depth=95, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=838; total time= 0.8s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=181; total time= 0.2s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=181; total time= 0.2s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=181; total time= 0.1s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=181; total time= 0.2s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=181; total time= 0.2s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=103; total time= 0.1s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=103; total time= 0.1s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=103; total time= 0.1s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=103; total time= 0.1s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=103; total time= 0.1s [CV] END bootstrap=False, max_depth=250, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=100; total time= 0.1s [CV] END bootstrap=False, max_depth=250, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=100; total time= 0.1s [CV] END bootstrap=False, max_depth=250, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=100; total time= 0.1s [CV] END bootstrap=False, max_depth=250, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=100; total time= 0.1s [CV] END bootstrap=False, max_depth=250, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=100; total time= 0.1s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=319; total time= 0.5s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=319; total time= 0.4s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=319; total time= 0.4s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=319; total time= 0.4s [CV] END bootstrap=False, max_depth=10, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=319; total time= 0.4s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=868; total time= 0.9s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=868; total time= 1.0s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=868; total time= 0.8s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=868; total time= 1.0s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=868; total time= 0.8s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=928; total time= 0.9s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=928; total time= 0.9s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=928; total time= 0.9s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=928; total time= 0.9s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=928; total time= 1.0s [CV] END bootstrap=True, max_depth=150, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=223; total time= 0.2s [CV] END bootstrap=True, max_depth=150, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=223; total time= 0.2s [CV] END bootstrap=True, max_depth=150, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=223; total time= 0.3s [CV] END bootstrap=True, max_depth=150, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=223; total time= 0.2s [CV] END bootstrap=True, max_depth=150, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=223; total time= 0.2s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=943; total time= 1.0s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=943; total time= 1.0s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=943; total time= 0.9s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=943; total time= 1.0s [CV] END bootstrap=True, max_depth=170, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=943; total time= 1.1s [CV] END bootstrap=False, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=808; total time= 0.7s [CV] END bootstrap=False, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=808; total time= 0.8s [CV] END bootstrap=False, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=808; total time= 0.6s [CV] END bootstrap=False, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=808; total time= 0.7s [CV] END bootstrap=False, max_depth=100, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=808; total time= 0.7s [CV] END bootstrap=False, max_depth=60, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=259; total time= 0.3s [CV] END bootstrap=False, max_depth=60, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=259; total time= 0.3s [CV] END bootstrap=False, max_depth=60, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=259; total time= 0.2s [CV] END bootstrap=False, max_depth=60, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=259; total time= 0.3s [CV] END bootstrap=False, max_depth=60, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=259; total time= 0.3s [CV] END bootstrap=False, max_depth=235, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=208; total time= 0.2s [CV] END bootstrap=False, max_depth=235, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=208; total time= 0.3s [CV] END bootstrap=False, max_depth=235, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=208; total time= 0.2s [CV] END bootstrap=False, max_depth=235, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=208; total time= 0.2s [CV] END bootstrap=False, max_depth=235, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=208; total time= 0.2s [CV] END bootstrap=False, max_depth=130, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=412; total time= 0.5s [CV] END bootstrap=False, max_depth=130, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=412; total time= 0.4s [CV] END bootstrap=False, max_depth=130, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=412; total time= 0.4s [CV] END bootstrap=False, max_depth=130, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=412; total time= 0.4s [CV] END bootstrap=False, max_depth=130, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=412; total time= 0.4s [CV] END bootstrap=False, max_depth=65, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=781; total time= 0.7s [CV] END bootstrap=False, max_depth=65, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=781; total time= 0.7s [CV] END bootstrap=False, max_depth=65, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=781; total time= 0.6s [CV] END bootstrap=False, max_depth=65, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=781; total time= 0.6s [CV] END bootstrap=False, max_depth=65, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=781; total time= 0.8s [CV] END bootstrap=False, max_depth=180, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=964; total time= 1.0s [CV] END bootstrap=False, max_depth=180, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=964; total time= 1.0s [CV] END bootstrap=False, max_depth=180, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=964; total time= 0.9s [CV] END bootstrap=False, max_depth=180, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=964; total time= 0.9s [CV] END bootstrap=False, max_depth=180, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=964; total time= 0.9s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=157; total time= 0.3s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=157; total time= 0.2s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=157; total time= 0.2s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=157; total time= 0.3s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=157; total time= 0.2s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=280; total time= 0.3s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=280; total time= 0.3s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=280; total time= 0.3s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=280; total time= 0.2s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=280; total time= 0.3s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=922; total time= 0.9s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=922; total time= 1.0s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=922; total time= 1.0s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=922; total time= 0.9s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=922; total time= 0.9s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=889; total time= 1.0s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=889; total time= 1.2s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=889; total time= 1.0s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=889; total time= 1.0s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=2, min_samples_split=2, n_estimators=889; total time= 1.0s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=553; total time= 0.5s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=553; total time= 0.6s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=553; total time= 0.6s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=553; total time= 0.5s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=553; total time= 0.6s [CV] END bootstrap=True, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=22; total time= 0.0s [CV] END bootstrap=True, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=22; total time= 0.0s [CV] END bootstrap=True, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=22; total time= 0.0s [CV] END bootstrap=True, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=22; total time= 0.0s [CV] END bootstrap=True, max_depth=30, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=22; total time= 0.0s [CV] END bootstrap=False, max_depth=130, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=481; total time= 0.7s [CV] END bootstrap=False, max_depth=130, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=481; total time= 0.6s [CV] END bootstrap=False, max_depth=130, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=481; total time= 0.6s [CV] END bootstrap=False, max_depth=130, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=481; total time= 0.6s [CV] END bootstrap=False, max_depth=130, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=481; total time= 0.7s [CV] END bootstrap=True, max_depth=175, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=151; total time= 0.2s [CV] END bootstrap=True, max_depth=175, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=151; total time= 0.2s [CV] END bootstrap=True, max_depth=175, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=151; total time= 0.2s [CV] END bootstrap=True, max_depth=175, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=151; total time= 0.2s [CV] END bootstrap=True, max_depth=175, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=151; total time= 0.2s [CV] END bootstrap=True, max_depth=160, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=343; total time= 0.4s [CV] END bootstrap=True, max_depth=160, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=343; total time= 0.5s [CV] END bootstrap=True, max_depth=160, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=343; total time= 0.4s [CV] END bootstrap=True, max_depth=160, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=343; total time= 0.5s [CV] END bootstrap=True, max_depth=160, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=343; total time= 0.4s [CV] END bootstrap=False, max_depth=190, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=112; total time= 0.1s [CV] END bootstrap=False, max_depth=190, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=112; total time= 0.1s [CV] END bootstrap=False, max_depth=190, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=112; total time= 0.1s [CV] END bootstrap=False, max_depth=190, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=112; total time= 0.1s [CV] END bootstrap=False, max_depth=190, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=112; total time= 0.2s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=586; total time= 0.7s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=586; total time= 0.7s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=586; total time= 0.6s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=586; total time= 0.6s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=586; total time= 0.7s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=508; total time= 0.4s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=508; total time= 0.4s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=508; total time= 0.5s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=508; total time= 0.4s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=508; total time= 0.3s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=781; total time= 0.6s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=781; total time= 0.8s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=781; total time= 0.7s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=781; total time= 0.6s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=781; total time= 0.6s [CV] END bootstrap=True, max_depth=80, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=574; total time= 0.7s [CV] END bootstrap=True, max_depth=80, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=574; total time= 0.6s [CV] END bootstrap=True, max_depth=80, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=574; total time= 0.6s [CV] END bootstrap=True, max_depth=80, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=574; total time= 0.6s [CV] END bootstrap=True, max_depth=80, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=574; total time= 0.6s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=757; total time= 0.6s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=757; total time= 0.9s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=757; total time= 0.6s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=757; total time= 0.6s [CV] END bootstrap=True, max_depth=55, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=757; total time= 0.5s [CV] END bootstrap=True, max_depth=250, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=382; total time= 0.3s [CV] END bootstrap=True, max_depth=250, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=382; total time= 0.3s [CV] END bootstrap=True, max_depth=250, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=382; total time= 0.3s [CV] END bootstrap=True, max_depth=250, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=382; total time= 0.3s [CV] END bootstrap=True, max_depth=250, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=382; total time= 0.3s [CV] END bootstrap=False, max_depth=200, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=61; total time= 0.0s [CV] END bootstrap=False, max_depth=200, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=61; total time= 0.1s [CV] END bootstrap=False, max_depth=200, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=61; total time= 0.1s [CV] END bootstrap=False, max_depth=200, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=61; total time= 0.1s [CV] END bootstrap=False, max_depth=200, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=61; total time= 0.0s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=625; total time= 0.8s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=625; total time= 0.8s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=625; total time= 0.7s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=625; total time= 0.8s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=625; total time= 0.7s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=817; total time= 0.7s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=817; total time= 0.8s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=817; total time= 0.8s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=817; total time= 0.9s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=817; total time= 0.8s [CV] END bootstrap=True, max_depth=15, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=13; total time= 0.0s [CV] END bootstrap=True, max_depth=15, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=13; total time= 0.0s [CV] END bootstrap=True, max_depth=15, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=13; total time= 0.0s [CV] END bootstrap=True, max_depth=15, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=13; total time= 0.0s [CV] END bootstrap=True, max_depth=15, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=13; total time= 0.0s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=481; total time= 0.6s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=481; total time= 0.6s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=481; total time= 0.6s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=481; total time= 0.6s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=481; total time= 0.6s [CV] END bootstrap=False, max_depth=30, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=19; total time= 0.0s [CV] END bootstrap=False, max_depth=30, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=19; total time= 0.0s [CV] END bootstrap=False, max_depth=30, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=19; total time= 0.0s [CV] END bootstrap=False, max_depth=30, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=19; total time= 0.0s [CV] END bootstrap=False, max_depth=30, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=19; total time= 0.0s [CV] END bootstrap=False, max_depth=110, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=19; total time= 0.0s [CV] END bootstrap=False, max_depth=110, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=19; total time= 0.0s [CV] END bootstrap=False, max_depth=110, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=19; total time= 0.0s [CV] END bootstrap=False, max_depth=110, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=19; total time= 0.0s [CV] END bootstrap=False, max_depth=110, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=19; total time= 0.0s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=115; total time= 0.1s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=115; total time= 0.1s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=115; total time= 0.2s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=115; total time= 0.2s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=115; total time= 0.1s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=763; total time= 0.7s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=763; total time= 0.7s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=763; total time= 0.6s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=763; total time= 0.6s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=763; total time= 0.6s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=706; total time= 0.7s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=706; total time= 0.9s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=706; total time= 0.8s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=706; total time= 0.8s [CV] END bootstrap=False, max_depth=120, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=706; total time= 0.8s [CV] END bootstrap=False, max_depth=35, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=709; total time= 0.8s [CV] END bootstrap=False, max_depth=35, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=709; total time= 0.9s [CV] END bootstrap=False, max_depth=35, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=709; total time= 0.8s [CV] END bootstrap=False, max_depth=35, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=709; total time= 0.8s [CV] END bootstrap=False, max_depth=35, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=709; total time= 0.7s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=196; total time= 0.2s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=196; total time= 0.3s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=196; total time= 0.2s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=196; total time= 0.2s [CV] END bootstrap=True, max_depth=130, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=196; total time= 0.2s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=31; total time= 0.0s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=31; total time= 0.0s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=31; total time= 0.0s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=31; total time= 0.0s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=31; total time= 0.0s [CV] END bootstrap=True, max_depth=80, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=154; total time= 0.3s [CV] END bootstrap=True, max_depth=80, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=154; total time= 0.2s [CV] END bootstrap=True, max_depth=80, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=154; total time= 0.2s [CV] END bootstrap=True, max_depth=80, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=154; total time= 0.2s [CV] END bootstrap=True, max_depth=80, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=154; total time= 0.2s [CV] END bootstrap=False, max_depth=215, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=649; total time= 0.8s [CV] END bootstrap=False, max_depth=215, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=649; total time= 0.8s [CV] END bootstrap=False, max_depth=215, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=649; total time= 0.9s [CV] END bootstrap=False, max_depth=215, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=649; total time= 0.8s [CV] END bootstrap=False, max_depth=215, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=649; total time= 0.8s [CV] END bootstrap=False, max_depth=150, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=376; total time= 0.3s [CV] END bootstrap=False, max_depth=150, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=376; total time= 0.3s [CV] END bootstrap=False, max_depth=150, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=376; total time= 0.4s [CV] END bootstrap=False, max_depth=150, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=376; total time= 0.3s [CV] END bootstrap=False, max_depth=150, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=376; total time= 0.4s [CV] END bootstrap=True, max_depth=160, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=628; total time= 0.8s [CV] END bootstrap=True, max_depth=160, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=628; total time= 0.7s [CV] END bootstrap=True, max_depth=160, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=628; total time= 0.8s [CV] END bootstrap=True, max_depth=160, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=628; total time= 0.9s [CV] END bootstrap=True, max_depth=160, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=628; total time= 0.7s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=802; total time= 0.8s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=802; total time= 0.8s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=802; total time= 0.7s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=802; total time= 0.7s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=802; total time= 0.7s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=829; total time= 0.9s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=829; total time= 0.8s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=829; total time= 0.9s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=829; total time= 0.8s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=829; total time= 0.9s [CV] END bootstrap=False, max_depth=45, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=130; total time= 0.1s [CV] END bootstrap=False, max_depth=45, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=130; total time= 0.1s [CV] END bootstrap=False, max_depth=45, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=130; total time= 0.2s [CV] END bootstrap=False, max_depth=45, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=130; total time= 0.2s [CV] END bootstrap=False, max_depth=45, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=130; total time= 0.1s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=439; total time= 0.5s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=439; total time= 0.4s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=439; total time= 0.5s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=439; total time= 0.4s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=439; total time= 0.5s [CV] END bootstrap=True, max_depth=95, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=394; total time= 0.3s [CV] END bootstrap=True, max_depth=95, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=394; total time= 0.4s [CV] END bootstrap=True, max_depth=95, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=394; total time= 0.3s [CV] END bootstrap=True, max_depth=95, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=394; total time= 0.4s [CV] END bootstrap=True, max_depth=95, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=394; total time= 0.3s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=454; total time= 0.6s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=454; total time= 0.5s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=454; total time= 0.5s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=454; total time= 0.6s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=454; total time= 0.5s [CV] END bootstrap=True, max_depth=30, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=622; total time= 0.8s [CV] END bootstrap=True, max_depth=30, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=622; total time= 0.8s [CV] END bootstrap=True, max_depth=30, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=622; total time= 0.8s [CV] END bootstrap=True, max_depth=30, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=622; total time= 0.8s [CV] END bootstrap=True, max_depth=30, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=622; total time= 0.7s [CV] END bootstrap=False, max_depth=200, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=619; total time= 0.8s [CV] END bootstrap=False, max_depth=200, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=619; total time= 0.6s [CV] END bootstrap=False, max_depth=200, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=619; total time= 0.8s [CV] END bootstrap=False, max_depth=200, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=619; total time= 0.9s [CV] END bootstrap=False, max_depth=200, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=619; total time= 0.8s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=799; total time= 0.7s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=799; total time= 0.8s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=799; total time= 0.6s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=799; total time= 0.7s [CV] END bootstrap=True, max_depth=105, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=799; total time= 0.8s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=742; total time= 0.7s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=742; total time= 0.8s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=742; total time= 0.7s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=742; total time= 0.8s [CV] END bootstrap=True, max_depth=250, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=742; total time= 0.9s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=139; total time= 0.1s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=139; total time= 0.1s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=139; total time= 0.2s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=139; total time= 0.2s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=139; total time= 0.2s [CV] END bootstrap=True, max_depth=25, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=457; total time= 0.5s [CV] END bootstrap=True, max_depth=25, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=457; total time= 0.6s [CV] END bootstrap=True, max_depth=25, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=457; total time= 0.7s [CV] END bootstrap=True, max_depth=25, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=457; total time= 0.5s [CV] END bootstrap=True, max_depth=25, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=457; total time= 0.5s [CV] END bootstrap=True, max_depth=215, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=703; total time= 0.8s [CV] END bootstrap=True, max_depth=215, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=703; total time= 0.8s [CV] END bootstrap=True, max_depth=215, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=703; total time= 0.7s [CV] END bootstrap=True, max_depth=215, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=703; total time= 0.8s [CV] END bootstrap=True, max_depth=215, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=703; total time= 0.8s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=277; total time= 0.3s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=277; total time= 0.3s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=277; total time= 0.3s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=277; total time= 0.3s [CV] END bootstrap=True, max_depth=65, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=277; total time= 0.3s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=469; total time= 0.6s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=469; total time= 0.6s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=469; total time= 0.5s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=469; total time= 0.5s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=469; total time= 0.7s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=856; total time= 1.0s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=856; total time= 0.9s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=856; total time= 0.9s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=856; total time= 0.9s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=856; total time= 1.0s [CV] END bootstrap=True, max_depth=185, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=934; total time= 1.0s [CV] END bootstrap=True, max_depth=185, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=934; total time= 1.0s [CV] END bootstrap=True, max_depth=185, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=934; total time= 1.0s [CV] END bootstrap=True, max_depth=185, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=934; total time= 1.0s [CV] END bootstrap=True, max_depth=185, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=934; total time= 0.9s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=916; total time= 1.2s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=916; total time= 1.1s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=916; total time= 0.9s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=916; total time= 0.9s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=916; total time= 0.9s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=556; total time= 0.7s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=556; total time= 0.6s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=556; total time= 0.5s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=556; total time= 0.7s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=556; total time= 0.5s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=904; total time= 1.0s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=904; total time= 1.0s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=904; total time= 1.1s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=904; total time= 1.0s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=904; total time= 1.1s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=940; total time= 1.0s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=940; total time= 1.0s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=940; total time= 1.0s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=940; total time= 0.9s [CV] END bootstrap=True, max_depth=75, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=940; total time= 1.0s [CV] END bootstrap=True, max_depth=240, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=385; total time= 0.3s [CV] END bootstrap=True, max_depth=240, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=385; total time= 0.4s [CV] END bootstrap=True, max_depth=240, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=385; total time= 0.3s [CV] END bootstrap=True, max_depth=240, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=385; total time= 0.4s [CV] END bootstrap=True, max_depth=240, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=385; total time= 0.3s [CV] END bootstrap=False, max_depth=50, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=271; total time= 0.3s [CV] END bootstrap=False, max_depth=50, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=271; total time= 0.3s [CV] END bootstrap=False, max_depth=50, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=271; total time= 0.2s [CV] END bootstrap=False, max_depth=50, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=271; total time= 0.3s [CV] END bootstrap=False, max_depth=50, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=271; total time= 0.3s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=283; total time= 0.3s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=283; total time= 0.3s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=283; total time= 0.3s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=283; total time= 0.4s [CV] END bootstrap=True, max_depth=145, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=283; total time= 0.3s [CV] END bootstrap=True, max_depth=50, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=796; total time= 0.7s [CV] END bootstrap=True, max_depth=50, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=796; total time= 0.7s [CV] END bootstrap=True, max_depth=50, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=796; total time= 0.7s [CV] END bootstrap=True, max_depth=50, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=796; total time= 0.6s [CV] END bootstrap=True, max_depth=50, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=796; total time= 0.8s [CV] END bootstrap=False, max_depth=180, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=244; total time= 0.2s [CV] END bootstrap=False, max_depth=180, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=244; total time= 0.2s [CV] END bootstrap=False, max_depth=180, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=244; total time= 0.2s [CV] END bootstrap=False, max_depth=180, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=244; total time= 0.2s [CV] END bootstrap=False, max_depth=180, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=244; total time= 0.2s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=787; total time= 0.6s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=787; total time= 0.7s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=787; total time= 0.6s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=787; total time= 0.6s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=787; total time= 0.7s [CV] END bootstrap=False, max_depth=210, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=337; total time= 0.7s [CV] END bootstrap=False, max_depth=210, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=337; total time= 0.5s [CV] END bootstrap=False, max_depth=210, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=337; total time= 0.5s [CV] END bootstrap=False, max_depth=210, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=337; total time= 0.5s [CV] END bootstrap=False, max_depth=210, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=337; total time= 0.6s [CV] END bootstrap=False, max_depth=255, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=697; total time= 0.6s [CV] END bootstrap=False, max_depth=255, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=697; total time= 0.7s [CV] END bootstrap=False, max_depth=255, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=697; total time= 0.8s [CV] END bootstrap=False, max_depth=255, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=697; total time= 0.7s [CV] END bootstrap=False, max_depth=255, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=697; total time= 0.8s [CV] END bootstrap=True, max_depth=125, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=598; total time= 0.7s [CV] END bootstrap=True, max_depth=125, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=598; total time= 0.7s [CV] END bootstrap=True, max_depth=125, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=598; total time= 0.7s [CV] END bootstrap=True, max_depth=125, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=598; total time= 0.6s [CV] END bootstrap=True, max_depth=125, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=598; total time= 0.6s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=586; total time= 0.6s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=586; total time= 0.8s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=586; total time= 0.9s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=586; total time= 0.8s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=586; total time= 0.8s [CV] END bootstrap=True, max_depth=185, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=733; total time= 1.0s [CV] END bootstrap=True, max_depth=185, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=733; total time= 1.1s [CV] END bootstrap=True, max_depth=185, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=733; total time= 1.0s [CV] END bootstrap=True, max_depth=185, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=733; total time= 1.0s [CV] END bootstrap=True, max_depth=185, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=733; total time= 1.1s [CV] END bootstrap=True, max_depth=180, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=442; total time= 0.5s [CV] END bootstrap=True, max_depth=180, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=442; total time= 0.7s [CV] END bootstrap=True, max_depth=180, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=442; total time= 0.6s [CV] END bootstrap=True, max_depth=180, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=442; total time= 0.6s [CV] END bootstrap=True, max_depth=180, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=442; total time= 0.6s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=181; total time= 0.2s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=181; total time= 0.1s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=181; total time= 0.1s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=181; total time= 0.1s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=181; total time= 0.1s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=214; total time= 0.1s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=214; total time= 0.1s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=214; total time= 0.1s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=214; total time= 0.1s [CV] END bootstrap=True, max_depth=205, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=214; total time= 0.1s [CV] END bootstrap=False, max_depth=115, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=952; total time= 0.5s [CV] END bootstrap=False, max_depth=115, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=952; total time= 0.5s [CV] END bootstrap=False, max_depth=115, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=952; total time= 0.5s [CV] END bootstrap=False, max_depth=115, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=952; total time= 0.5s [CV] END bootstrap=False, max_depth=115, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=952; total time= 0.5s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=502; total time= 0.4s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=502; total time= 0.5s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=502; total time= 0.4s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=502; total time= 0.4s [CV] END bootstrap=True, max_depth=185, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=502; total time= 0.4s [CV] END bootstrap=False, max_depth=40, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=127; total time= 0.1s [CV] END bootstrap=False, max_depth=40, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=127; total time= 0.1s [CV] END bootstrap=False, max_depth=40, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=127; total time= 0.1s [CV] END bootstrap=False, max_depth=40, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=127; total time= 0.2s [CV] END bootstrap=False, max_depth=40, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=127; total time= 0.1s [CV] END bootstrap=True, max_depth=15, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=43; total time= 0.0s [CV] END bootstrap=True, max_depth=15, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=43; total time= 0.0s [CV] END bootstrap=True, max_depth=15, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=43; total time= 0.0s [CV] END bootstrap=True, max_depth=15, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=43; total time= 0.0s [CV] END bootstrap=True, max_depth=15, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=43; total time= 0.0s [CV] END bootstrap=True, max_depth=250, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=166; total time= 0.2s [CV] END bootstrap=True, max_depth=250, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=166; total time= 0.2s [CV] END bootstrap=True, max_depth=250, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=166; total time= 0.2s [CV] END bootstrap=True, max_depth=250, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=166; total time= 0.2s [CV] END bootstrap=True, max_depth=250, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=166; total time= 0.3s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=256; total time= 0.3s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=256; total time= 0.2s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=256; total time= 0.2s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=256; total time= 0.2s [CV] END bootstrap=True, max_depth=200, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=256; total time= 0.3s [CV] END bootstrap=False, max_depth=70, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=160; total time= 0.2s [CV] END bootstrap=False, max_depth=70, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=160; total time= 0.3s [CV] END bootstrap=False, max_depth=70, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=160; total time= 0.4s [CV] END bootstrap=False, max_depth=70, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=160; total time= 0.2s [CV] END bootstrap=False, max_depth=70, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=160; total time= 0.3s [CV] END bootstrap=True, max_depth=210, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=889; total time= 1.0s [CV] END bootstrap=True, max_depth=210, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=889; total time= 1.0s [CV] END bootstrap=True, max_depth=210, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=889; total time= 0.9s [CV] END bootstrap=True, max_depth=210, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=889; total time= 1.0s [CV] END bootstrap=True, max_depth=210, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=889; total time= 1.1s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=745; total time= 0.6s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=745; total time= 0.7s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=745; total time= 0.7s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=745; total time= 0.8s [CV] END bootstrap=False, max_depth=105, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=745; total time= 0.7s [CV] END bootstrap=False, max_depth=20, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=820; total time= 0.8s [CV] END bootstrap=False, max_depth=20, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=820; total time= 0.7s [CV] END bootstrap=False, max_depth=20, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=820; total time= 0.8s [CV] END bootstrap=False, max_depth=20, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=820; total time= 0.8s [CV] END bootstrap=False, max_depth=20, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=820; total time= 0.8s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=52; total time= 0.1s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=52; total time= 0.0s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=52; total time= 0.0s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=52; total time= 0.1s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=52; total time= 0.0s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=739; total time= 0.7s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=739; total time= 0.7s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=739; total time= 0.8s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=739; total time= 0.7s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=739; total time= 0.8s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=280; total time= 0.3s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=280; total time= 0.3s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=280; total time= 0.3s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=280; total time= 0.3s [CV] END bootstrap=False, max_depth=145, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=280; total time= 0.3s [CV] END bootstrap=True, max_depth=225, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=301; total time= 0.3s [CV] END bootstrap=True, max_depth=225, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=301; total time= 0.4s [CV] END bootstrap=True, max_depth=225, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=301; total time= 0.4s [CV] END bootstrap=True, max_depth=225, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=301; total time= 0.4s [CV] END bootstrap=True, max_depth=225, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=301; total time= 0.4s [CV] END bootstrap=True, max_depth=115, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=49; total time= 0.0s [CV] END bootstrap=True, max_depth=115, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=49; total time= 0.0s [CV] END bootstrap=True, max_depth=115, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=49; total time= 0.0s [CV] END bootstrap=True, max_depth=115, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=49; total time= 0.0s [CV] END bootstrap=True, max_depth=115, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=49; total time= 0.0s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=568; total time= 0.7s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=568; total time= 0.7s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=568; total time= 0.6s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=568; total time= 0.6s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=568; total time= 0.6s [CV] END bootstrap=True, max_depth=210, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=757; total time= 0.5s [CV] END bootstrap=True, max_depth=210, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=757; total time= 0.5s [CV] END bootstrap=True, max_depth=210, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=757; total time= 0.6s [CV] END bootstrap=True, max_depth=210, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=757; total time= 0.6s [CV] END bootstrap=True, max_depth=210, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=757; total time= 0.7s [CV] END bootstrap=False, max_depth=65, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=649; total time= 0.8s [CV] END bootstrap=False, max_depth=65, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=649; total time= 0.8s [CV] END bootstrap=False, max_depth=65, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=649; total time= 0.9s [CV] END bootstrap=False, max_depth=65, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=649; total time= 0.9s [CV] END bootstrap=False, max_depth=65, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=649; total time= 0.8s [CV] END bootstrap=False, max_depth=160, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=760; total time= 0.7s [CV] END bootstrap=False, max_depth=160, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=760; total time= 0.6s [CV] END bootstrap=False, max_depth=160, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=760; total time= 0.6s [CV] END bootstrap=False, max_depth=160, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=760; total time= 0.6s [CV] END bootstrap=False, max_depth=160, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=760; total time= 0.7s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=442; total time= 0.5s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=442; total time= 0.5s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=442; total time= 0.5s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=442; total time= 0.6s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=442; total time= 0.5s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=190; total time= 0.1s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=190; total time= 0.1s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=190; total time= 0.2s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=190; total time= 0.2s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=190; total time= 0.2s [CV] END bootstrap=False, max_depth=55, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=37; total time= 0.0s [CV] END bootstrap=False, max_depth=55, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=37; total time= 0.0s [CV] END bootstrap=False, max_depth=55, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=37; total time= 0.0s [CV] END bootstrap=False, max_depth=55, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=37; total time= 0.0s [CV] END bootstrap=False, max_depth=55, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=37; total time= 0.0s [CV] END bootstrap=True, max_depth=210, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=436; total time= 0.4s [CV] END bootstrap=True, max_depth=210, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=436; total time= 0.6s [CV] END bootstrap=True, max_depth=210, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=436; total time= 0.5s [CV] END bootstrap=True, max_depth=210, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=436; total time= 0.5s [CV] END bootstrap=True, max_depth=210, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=436; total time= 0.5s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=931; total time= 0.9s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=931; total time= 1.0s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=931; total time= 1.0s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=931; total time= 1.0s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=931; total time= 0.9s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=481; total time= 0.6s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=481; total time= 0.6s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=481; total time= 0.6s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=481; total time= 0.7s [CV] END bootstrap=True, max_depth=225, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=481; total time= 0.7s [CV] END bootstrap=True, max_depth=250, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=916; total time= 1.1s [CV] END bootstrap=True, max_depth=250, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=916; total time= 1.0s [CV] END bootstrap=True, max_depth=250, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=916; total time= 1.0s [CV] END bootstrap=True, max_depth=250, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=916; total time= 0.9s [CV] END bootstrap=True, max_depth=250, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=916; total time= 0.9s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=610; total time= 0.7s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=610; total time= 0.7s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=610; total time= 0.8s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=610; total time= 0.6s [CV] END bootstrap=True, max_depth=120, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=610; total time= 0.6s [CV] END bootstrap=True, max_depth=80, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=745; total time= 0.8s [CV] END bootstrap=True, max_depth=80, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=745; total time= 0.8s [CV] END bootstrap=True, max_depth=80, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=745; total time= 0.7s [CV] END bootstrap=True, max_depth=80, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=745; total time= 0.8s [CV] END bootstrap=True, max_depth=80, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=745; total time= 0.8s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=916; total time= 1.0s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=916; total time= 1.0s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=916; total time= 0.9s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=916; total time= 0.9s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=916; total time= 1.0s [CV] END bootstrap=True, max_depth=135, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=196; total time= 0.2s [CV] END bootstrap=True, max_depth=135, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=196; total time= 0.1s [CV] END bootstrap=True, max_depth=135, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=196; total time= 0.4s [CV] END bootstrap=True, max_depth=135, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=196; total time= 0.2s [CV] END bootstrap=True, max_depth=135, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=196; total time= 0.2s [CV] END bootstrap=True, max_depth=245, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=769; total time= 0.7s [CV] END bootstrap=True, max_depth=245, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=769; total time= 0.6s [CV] END bootstrap=True, max_depth=245, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=769; total time= 0.7s [CV] END bootstrap=True, max_depth=245, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=769; total time= 0.6s [CV] END bootstrap=True, max_depth=245, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=769; total time= 0.6s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=604; total time= 0.8s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=604; total time= 0.7s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=604; total time= 0.6s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=604; total time= 0.7s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=604; total time= 0.6s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=997; total time= 1.0s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=997; total time= 1.0s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=997; total time= 0.9s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=997; total time= 0.9s [CV] END bootstrap=True, max_depth=215, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=997; total time= 0.9s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=664; total time= 0.7s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=664; total time= 0.8s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=664; total time= 0.8s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=664; total time= 0.6s [CV] END bootstrap=False, max_depth=245, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=664; total time= 0.7s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=985; total time= 1.1s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=985; total time= 1.1s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=985; total time= 1.0s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=985; total time= 1.0s [CV] END bootstrap=False, max_depth=50, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=985; total time= 1.1s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=259; total time= 0.3s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=259; total time= 0.2s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=259; total time= 0.2s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=259; total time= 0.3s [CV] END bootstrap=False, max_depth=80, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=259; total time= 0.3s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=568; total time= 0.5s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=568; total time= 0.6s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=568; total time= 0.6s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=568; total time= 0.6s [CV] END bootstrap=False, max_depth=125, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=568; total time= 0.5s [CV] END bootstrap=True, max_depth=40, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=361; total time= 0.5s [CV] END bootstrap=True, max_depth=40, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=361; total time= 0.4s [CV] END bootstrap=True, max_depth=40, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=361; total time= 0.5s [CV] END bootstrap=True, max_depth=40, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=361; total time= 0.4s [CV] END bootstrap=True, max_depth=40, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=361; total time= 0.5s [CV] END bootstrap=False, max_depth=220, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=754; total time= 0.7s [CV] END bootstrap=False, max_depth=220, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=754; total time= 0.6s [CV] END bootstrap=False, max_depth=220, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=754; total time= 0.5s [CV] END bootstrap=False, max_depth=220, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=754; total time= 0.6s [CV] END bootstrap=False, max_depth=220, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=754; total time= 0.7s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=82; total time= 0.0s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=82; total time= 0.1s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=82; total time= 0.1s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=82; total time= 0.1s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=82; total time= 0.1s [CV] END bootstrap=False, max_depth=225, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=961; total time= 1.0s [CV] END bootstrap=False, max_depth=225, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=961; total time= 1.0s [CV] END bootstrap=False, max_depth=225, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=961; total time= 0.9s [CV] END bootstrap=False, max_depth=225, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=961; total time= 0.9s [CV] END bootstrap=False, max_depth=225, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=961; total time= 1.0s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=907; total time= 1.1s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=907; total time= 1.1s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=907; total time= 1.1s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=907; total time= 1.2s [CV] END bootstrap=True, max_depth=110, max_features=sqrt, min_samples_leaf=4, min_samples_split=50, n_estimators=907; total time= 1.1s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=916; total time= 1.1s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=916; total time= 1.0s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=916; total time= 0.9s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=916; total time= 1.2s [CV] END bootstrap=False, max_depth=20, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=916; total time= 1.1s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=400; total time= 0.4s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=400; total time= 0.4s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=400; total time= 0.3s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=400; total time= 0.4s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=400; total time= 0.3s [CV] END bootstrap=True, max_depth=150, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=571; total time= 0.6s [CV] END bootstrap=True, max_depth=150, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=571; total time= 0.6s [CV] END bootstrap=True, max_depth=150, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=571; total time= 0.6s [CV] END bootstrap=True, max_depth=150, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=571; total time= 0.7s [CV] END bootstrap=True, max_depth=150, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=571; total time= 0.6s [CV] END bootstrap=False, max_depth=210, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=334; total time= 0.5s [CV] END bootstrap=False, max_depth=210, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=334; total time= 0.4s [CV] END bootstrap=False, max_depth=210, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=334; total time= 0.5s [CV] END bootstrap=False, max_depth=210, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=334; total time= 0.4s [CV] END bootstrap=False, max_depth=210, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=334; total time= 0.4s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=196; total time= 0.2s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=196; total time= 0.1s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=196; total time= 0.2s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=196; total time= 0.2s [CV] END bootstrap=True, max_depth=20, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=196; total time= 0.2s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=103; total time= 0.1s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=103; total time= 0.1s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=103; total time= 0.1s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=103; total time= 0.2s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=103; total time= 0.1s [CV] END bootstrap=True, max_depth=130, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=145; total time= 0.2s [CV] END bootstrap=True, max_depth=130, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=145; total time= 0.2s [CV] END bootstrap=True, max_depth=130, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=145; total time= 0.2s [CV] END bootstrap=True, max_depth=130, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=145; total time= 0.2s [CV] END bootstrap=True, max_depth=130, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=145; total time= 0.2s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=106; total time= 0.2s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=106; total time= 0.1s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=106; total time= 0.1s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=106; total time= 0.1s [CV] END bootstrap=False, max_depth=175, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=106; total time= 0.1s [CV] END bootstrap=True, max_depth=200, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=928; total time= 0.9s [CV] END bootstrap=True, max_depth=200, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=928; total time= 0.8s [CV] END bootstrap=True, max_depth=200, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=928; total time= 1.0s [CV] END bootstrap=True, max_depth=200, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=928; total time= 0.9s [CV] END bootstrap=True, max_depth=200, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=928; total time= 1.0s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=262; total time= 0.2s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=262; total time= 0.3s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=262; total time= 0.2s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=262; total time= 0.2s [CV] END bootstrap=False, max_depth=215, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=262; total time= 0.2s [CV] END bootstrap=True, max_depth=135, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=910; total time= 1.0s [CV] END bootstrap=True, max_depth=135, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=910; total time= 1.1s [CV] END bootstrap=True, max_depth=135, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=910; total time= 1.1s [CV] END bootstrap=True, max_depth=135, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=910; total time= 1.0s [CV] END bootstrap=True, max_depth=135, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=910; total time= 1.0s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=604; total time= 0.7s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=604; total time= 0.7s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=604; total time= 0.7s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=604; total time= 0.6s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=2, min_samples_split=2, n_estimators=604; total time= 0.9s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=784; total time= 0.7s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=784; total time= 0.7s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=784; total time= 0.7s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=784; total time= 0.6s [CV] END bootstrap=False, max_depth=230, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=784; total time= 0.8s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=919; total time= 1.1s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=919; total time= 1.0s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=919; total time= 1.2s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=919; total time= 0.9s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=919; total time= 0.9s [CV] END bootstrap=False, max_depth=30, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=280; total time= 0.3s [CV] END bootstrap=False, max_depth=30, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=280; total time= 0.3s [CV] END bootstrap=False, max_depth=30, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=280; total time= 0.3s [CV] END bootstrap=False, max_depth=30, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=280; total time= 0.3s [CV] END bootstrap=False, max_depth=30, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=280; total time= 0.3s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=373; total time= 0.3s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=373; total time= 0.3s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=373; total time= 0.4s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=373; total time= 0.3s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=373; total time= 0.3s [CV] END bootstrap=False, max_depth=180, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=793; total time= 0.7s [CV] END bootstrap=False, max_depth=180, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=793; total time= 0.7s [CV] END bootstrap=False, max_depth=180, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=793; total time= 0.7s [CV] END bootstrap=False, max_depth=180, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=793; total time= 0.7s [CV] END bootstrap=False, max_depth=180, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=793; total time= 0.8s [CV] END bootstrap=True, max_depth=150, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=268; total time= 0.3s [CV] END bootstrap=True, max_depth=150, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=268; total time= 0.3s [CV] END bootstrap=True, max_depth=150, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=268; total time= 0.2s [CV] END bootstrap=True, max_depth=150, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=268; total time= 0.3s [CV] END bootstrap=True, max_depth=150, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=268; total time= 0.2s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=625; total time= 0.8s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=625; total time= 0.9s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=625; total time= 0.7s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=625; total time= 0.7s [CV] END bootstrap=False, max_depth=15, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=625; total time= 0.6s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=502; total time= 0.4s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=502; total time= 0.7s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=502; total time= 0.4s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=502; total time= 0.4s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=502; total time= 0.6s [CV] END bootstrap=True, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=655; total time= 1.0s [CV] END bootstrap=True, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=655; total time= 0.9s [CV] END bootstrap=True, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=655; total time= 0.8s [CV] END bootstrap=True, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=655; total time= 0.9s [CV] END bootstrap=True, max_depth=190, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=655; total time= 0.9s [CV] END bootstrap=True, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=217; total time= 0.3s [CV] END bootstrap=True, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=217; total time= 0.2s [CV] END bootstrap=True, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=217; total time= 0.2s [CV] END bootstrap=True, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=217; total time= 0.3s [CV] END bootstrap=True, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=10, n_estimators=217; total time= 0.3s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=211; total time= 0.2s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=211; total time= 0.2s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=211; total time= 0.2s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=211; total time= 0.3s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=211; total time= 0.2s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=472; total time= 0.6s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=472; total time= 0.6s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=472; total time= 0.6s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=472; total time= 0.6s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=10, min_samples_split=2, n_estimators=472; total time= 0.6s [CV] END bootstrap=True, max_depth=60, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=718; total time= 0.8s [CV] END bootstrap=True, max_depth=60, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=718; total time= 0.8s [CV] END bootstrap=True, max_depth=60, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=718; total time= 0.8s [CV] END bootstrap=True, max_depth=60, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=718; total time= 0.8s [CV] END bootstrap=True, max_depth=60, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=718; total time= 0.8s [CV] END bootstrap=False, max_depth=155, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=946; total time= 1.0s [CV] END bootstrap=False, max_depth=155, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=946; total time= 1.0s [CV] END bootstrap=False, max_depth=155, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=946; total time= 0.9s [CV] END bootstrap=False, max_depth=155, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=946; total time= 1.0s [CV] END bootstrap=False, max_depth=155, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=946; total time= 1.0s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=412; total time= 0.4s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=412; total time= 0.4s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=412; total time= 0.4s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=412; total time= 0.5s [CV] END bootstrap=True, max_depth=70, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=412; total time= 0.5s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=430; total time= 0.5s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=430; total time= 0.5s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=430; total time= 0.4s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=430; total time= 0.5s [CV] END bootstrap=False, max_depth=170, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=430; total time= 0.5s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=265; total time= 0.3s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=265; total time= 0.2s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=265; total time= 0.3s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=265; total time= 0.2s [CV] END bootstrap=True, max_depth=65, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=265; total time= 0.2s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=172; total time= 0.3s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=172; total time= 0.2s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=172; total time= 0.2s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=172; total time= 0.3s [CV] END bootstrap=True, max_depth=90, max_features=sqrt, min_samples_leaf=20, min_samples_split=50, n_estimators=172; total time= 0.2s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=439; total time= 0.4s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=439; total time= 0.5s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=439; total time= 0.5s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=439; total time= 0.5s [CV] END bootstrap=True, max_depth=35, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=439; total time= 0.5s [CV] END bootstrap=True, max_depth=140, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=685; total time= 0.8s [CV] END bootstrap=True, max_depth=140, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=685; total time= 0.7s [CV] END bootstrap=True, max_depth=140, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=685; total time= 0.8s [CV] END bootstrap=True, max_depth=140, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=685; total time= 0.8s [CV] END bootstrap=True, max_depth=140, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=685; total time= 0.8s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=661; total time= 0.7s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=661; total time= 0.8s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=661; total time= 0.7s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=661; total time= 0.7s [CV] END bootstrap=False, max_depth=255, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=661; total time= 0.7s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=721; total time= 0.8s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=721; total time= 0.9s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=721; total time= 0.7s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=721; total time= 1.0s [CV] END bootstrap=False, max_depth=135, max_features=sqrt, min_samples_leaf=20, min_samples_split=2, n_estimators=721; total time= 0.8s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=862; total time= 0.9s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=862; total time= 1.0s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=862; total time= 0.9s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=862; total time= 1.0s [CV] END bootstrap=True, max_depth=240, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=862; total time= 0.9s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=97; total time= 0.1s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=97; total time= 0.1s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=97; total time= 0.1s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=97; total time= 0.1s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=97; total time= 0.1s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=715; total time= 0.9s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=715; total time= 0.8s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=715; total time= 0.8s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=715; total time= 0.8s [CV] END bootstrap=True, max_depth=90, max_features=auto, min_samples_leaf=2, min_samples_split=50, n_estimators=715; total time= 0.9s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=913; total time= 1.0s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=913; total time= 1.0s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=913; total time= 1.0s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=913; total time= 0.9s [CV] END bootstrap=False, max_depth=220, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=913; total time= 0.9s [CV] END bootstrap=False, max_depth=75, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=778; total time= 0.7s [CV] END bootstrap=False, max_depth=75, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=778; total time= 0.7s [CV] END bootstrap=False, max_depth=75, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=778; total time= 0.7s [CV] END bootstrap=False, max_depth=75, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=778; total time= 0.7s [CV] END bootstrap=False, max_depth=75, max_features=auto, min_samples_leaf=20, min_samples_split=10, n_estimators=778; total time= 0.7s [CV] END bootstrap=True, max_depth=50, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=838; total time= 0.9s [CV] END bootstrap=True, max_depth=50, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=838; total time= 0.9s [CV] END bootstrap=True, max_depth=50, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=838; total time= 0.8s [CV] END bootstrap=True, max_depth=50, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=838; total time= 0.9s [CV] END bootstrap=True, max_depth=50, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=838; total time= 0.8s [CV] END bootstrap=True, max_depth=210, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=862; total time= 0.9s [CV] END bootstrap=True, max_depth=210, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=862; total time= 1.0s [CV] END bootstrap=True, max_depth=210, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=862; total time= 1.1s [CV] END bootstrap=True, max_depth=210, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=862; total time= 0.8s [CV] END bootstrap=True, max_depth=210, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=862; total time= 0.9s [CV] END bootstrap=False, max_depth=205, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=175; total time= 0.2s [CV] END bootstrap=False, max_depth=205, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=175; total time= 0.2s [CV] END bootstrap=False, max_depth=205, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=175; total time= 0.2s [CV] END bootstrap=False, max_depth=205, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=175; total time= 0.2s [CV] END bootstrap=False, max_depth=205, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=175; total time= 0.3s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=502; total time= 0.4s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=502; total time= 0.5s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=502; total time= 0.7s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=502; total time= 0.5s [CV] END bootstrap=True, max_depth=160, max_features=sqrt, min_samples_leaf=10, min_samples_split=5, n_estimators=502; total time= 0.4s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=421; total time= 0.5s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=421; total time= 0.5s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=421; total time= 0.4s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=421; total time= 0.5s [CV] END bootstrap=True, max_depth=155, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=421; total time= 0.5s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=238; total time= 0.3s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=238; total time= 0.3s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=238; total time= 0.3s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=238; total time= 0.3s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=238; total time= 0.4s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=148; total time= 0.1s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=148; total time= 0.2s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=148; total time= 0.2s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=148; total time= 0.1s [CV] END bootstrap=True, max_depth=205, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=148; total time= 0.2s [CV] END bootstrap=False, max_depth=60, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=553; total time= 0.6s [CV] END bootstrap=False, max_depth=60, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=553; total time= 0.5s [CV] END bootstrap=False, max_depth=60, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=553; total time= 0.5s [CV] END bootstrap=False, max_depth=60, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=553; total time= 0.6s [CV] END bootstrap=False, max_depth=60, max_features=auto, min_samples_leaf=10, min_samples_split=50, n_estimators=553; total time= 0.6s [CV] END bootstrap=True, max_depth=180, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=628; total time= 0.8s [CV] END bootstrap=True, max_depth=180, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=628; total time= 0.8s [CV] END bootstrap=True, max_depth=180, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=628; total time= 0.7s [CV] END bootstrap=True, max_depth=180, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=628; total time= 0.8s [CV] END bootstrap=True, max_depth=180, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=628; total time= 0.8s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=946; total time= 1.0s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=946; total time= 1.0s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=946; total time= 1.0s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=946; total time= 0.9s [CV] END bootstrap=False, max_depth=105, max_features=auto, min_samples_leaf=4, min_samples_split=2, n_estimators=946; total time= 1.0s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=721; total time= 0.8s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=721; total time= 0.9s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=721; total time= 0.8s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=721; total time= 0.9s [CV] END bootstrap=False, max_depth=195, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=721; total time= 0.9s [CV] END bootstrap=True, max_depth=235, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=475; total time= 0.6s [CV] END bootstrap=True, max_depth=235, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=475; total time= 0.5s [CV] END bootstrap=True, max_depth=235, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=475; total time= 0.6s [CV] END bootstrap=True, max_depth=235, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=475; total time= 0.6s [CV] END bootstrap=True, max_depth=235, max_features=sqrt, min_samples_leaf=20, min_samples_split=5, n_estimators=475; total time= 0.6s [CV] END bootstrap=True, max_depth=245, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=400; total time= 0.3s [CV] END bootstrap=True, max_depth=245, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=400; total time= 0.3s [CV] END bootstrap=True, max_depth=245, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=400; total time= 0.4s [CV] END bootstrap=True, max_depth=245, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=400; total time= 0.4s [CV] END bootstrap=True, max_depth=245, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=400; total time= 0.4s [CV] END bootstrap=True, max_depth=25, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=931; total time= 0.9s [CV] END bootstrap=True, max_depth=25, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=931; total time= 0.9s [CV] END bootstrap=True, max_depth=25, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=931; total time= 0.9s [CV] END bootstrap=True, max_depth=25, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=931; total time= 1.0s [CV] END bootstrap=True, max_depth=25, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=931; total time= 0.9s [CV] END bootstrap=False, max_depth=30, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=307; total time= 0.3s [CV] END bootstrap=False, max_depth=30, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=307; total time= 0.5s [CV] END bootstrap=False, max_depth=30, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=307; total time= 0.3s [CV] END bootstrap=False, max_depth=30, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=307; total time= 0.4s [CV] END bootstrap=False, max_depth=30, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=307; total time= 0.4s [CV] END bootstrap=False, max_depth=35, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=283; total time= 0.3s [CV] END bootstrap=False, max_depth=35, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=283; total time= 0.3s [CV] END bootstrap=False, max_depth=35, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=283; total time= 0.3s [CV] END bootstrap=False, max_depth=35, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=283; total time= 0.3s [CV] END bootstrap=False, max_depth=35, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=283; total time= 0.3s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=355; total time= 0.5s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=355; total time= 0.5s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=355; total time= 0.5s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=355; total time= 0.6s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=355; total time= 0.6s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=1000; total time= 1.0s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=1000; total time= 0.9s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=1000; total time= 0.9s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=1000; total time= 1.0s [CV] END bootstrap=False, max_depth=100, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=1000; total time= 1.0s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=610; total time= 0.8s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=610; total time= 0.8s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=610; total time= 0.7s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=610; total time= 0.7s [CV] END bootstrap=True, max_depth=100, max_features=sqrt, min_samples_leaf=2, min_samples_split=50, n_estimators=610; total time= 0.7s [CV] END bootstrap=True, max_depth=190, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=862; total time= 0.9s [CV] END bootstrap=True, max_depth=190, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=862; total time= 0.9s [CV] END bootstrap=True, max_depth=190, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=862; total time= 0.9s [CV] END bootstrap=True, max_depth=190, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=862; total time= 0.9s [CV] END bootstrap=True, max_depth=190, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=862; total time= 1.0s [CV] END bootstrap=True, max_depth=175, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=85; total time= 0.1s [CV] END bootstrap=True, max_depth=175, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=85; total time= 0.1s [CV] END bootstrap=True, max_depth=175, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=85; total time= 0.1s [CV] END bootstrap=True, max_depth=175, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=85; total time= 0.1s [CV] END bootstrap=True, max_depth=175, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=85; total time= 0.1s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=118; total time= 0.1s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=118; total time= 0.1s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=118; total time= 0.1s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=118; total time= 0.1s [CV] END bootstrap=True, max_depth=125, max_features=auto, min_samples_leaf=10, min_samples_split=5, n_estimators=118; total time= 0.1s [CV] END bootstrap=False, max_depth=65, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=928; total time= 0.9s [CV] END bootstrap=False, max_depth=65, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=928; total time= 0.9s [CV] END bootstrap=False, max_depth=65, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=928; total time= 1.0s [CV] END bootstrap=False, max_depth=65, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=928; total time= 1.0s [CV] END bootstrap=False, max_depth=65, max_features=auto, min_samples_leaf=4, min_samples_split=50, n_estimators=928; total time= 0.9s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=256; total time= 0.3s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=256; total time= 0.3s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=256; total time= 0.2s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=256; total time= 0.2s [CV] END bootstrap=False, max_depth=130, max_features=sqrt, min_samples_leaf=4, min_samples_split=5, n_estimators=256; total time= 0.2s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=460; total time= 0.4s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=460; total time= 0.6s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=460; total time= 0.6s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=460; total time= 0.5s [CV] END bootstrap=True, max_depth=220, max_features=auto, min_samples_leaf=2, min_samples_split=5, n_estimators=460; total time= 0.6s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=34; total time= 0.0s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=34; total time= 0.0s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=34; total time= 0.0s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=34; total time= 0.0s [CV] END bootstrap=True, max_depth=195, max_features=auto, min_samples_leaf=10, min_samples_split=10, n_estimators=34; total time= 0.0s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=118; total time= 0.1s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=118; total time= 0.2s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=118; total time= 0.1s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=118; total time= 0.1s [CV] END bootstrap=False, max_depth=140, max_features=sqrt, min_samples_leaf=2, min_samples_split=10, n_estimators=118; total time= 0.2s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=241; total time= 0.2s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=241; total time= 0.2s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=241; total time= 0.2s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=241; total time= 0.2s [CV] END bootstrap=True, max_depth=170, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=241; total time= 0.2s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=454; total time= 0.6s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=454; total time= 0.5s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=454; total time= 0.5s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=454; total time= 0.5s [CV] END bootstrap=False, max_depth=165, max_features=sqrt, min_samples_leaf=4, min_samples_split=2, n_estimators=454; total time= 0.5s [CV] END bootstrap=False, max_depth=220, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=298; total time= 0.4s [CV] END bootstrap=False, max_depth=220, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=298; total time= 0.3s [CV] END bootstrap=False, max_depth=220, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=298; total time= 0.4s [CV] END bootstrap=False, max_depth=220, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=298; total time= 0.4s [CV] END bootstrap=False, max_depth=220, max_features=auto, min_samples_leaf=20, min_samples_split=2, n_estimators=298; total time= 0.4s [CV] END bootstrap=True, max_depth=80, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=394; total time= 0.4s [CV] END bootstrap=True, max_depth=80, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=394; total time= 0.3s [CV] END bootstrap=True, max_depth=80, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=394; total time= 0.3s [CV] END bootstrap=True, max_depth=80, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=394; total time= 0.4s [CV] END bootstrap=True, max_depth=80, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=394; total time= 0.4s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=493; total time= 0.7s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=493; total time= 0.6s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=493; total time= 0.6s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=493; total time= 0.7s [CV] END bootstrap=True, max_depth=165, max_features=auto, min_samples_leaf=10, min_samples_split=2, n_estimators=493; total time= 0.7s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=343; total time= 0.5s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=343; total time= 0.5s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=343; total time= 0.5s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=343; total time= 0.4s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=343; total time= 0.5s [CV] END bootstrap=False, max_depth=215, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=367; total time= 0.6s [CV] END bootstrap=False, max_depth=215, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=367; total time= 0.5s [CV] END bootstrap=False, max_depth=215, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=367; total time= 0.6s [CV] END bootstrap=False, max_depth=215, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=367; total time= 0.5s [CV] END bootstrap=False, max_depth=215, max_features=auto, min_samples_leaf=4, min_samples_split=10, n_estimators=367; total time= 0.5s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=97; total time= 0.1s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=97; total time= 0.1s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=97; total time= 0.1s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=97; total time= 0.1s [CV] END bootstrap=False, max_depth=160, max_features=sqrt, min_samples_leaf=10, min_samples_split=50, n_estimators=97; total time= 0.1s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=736; total time= 0.8s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=736; total time= 0.8s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=736; total time= 0.7s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=736; total time= 0.8s [CV] END bootstrap=False, max_depth=85, max_features=sqrt, min_samples_leaf=2, min_samples_split=5, n_estimators=736; total time= 0.8s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=64; total time= 0.1s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=64; total time= 0.1s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=64; total time= 0.1s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=64; total time= 0.1s [CV] END bootstrap=True, max_depth=235, max_features=auto, min_samples_leaf=20, min_samples_split=50, n_estimators=64; total time= 0.1s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=478; total time= 0.6s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=478; total time= 0.6s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=478; total time= 0.6s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=478; total time= 0.5s [CV] END bootstrap=True, max_depth=155, max_features=auto, min_samples_leaf=4, min_samples_split=5, n_estimators=478; total time= 0.6s [CV] END bootstrap=False, max_depth=195, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=715; total time= 0.8s [CV] END bootstrap=False, max_depth=195, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=715; total time= 0.8s [CV] END bootstrap=False, max_depth=195, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=715; total time= 0.7s [CV] END bootstrap=False, max_depth=195, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=715; total time= 0.7s [CV] END bootstrap=False, max_depth=195, max_features=auto, min_samples_leaf=20, min_samples_split=5, n_estimators=715; total time= 0.8s [CV] END bootstrap=True, max_depth=255, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=790; total time= 0.7s [CV] END bootstrap=True, max_depth=255, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=790; total time= 0.7s [CV] END bootstrap=True, max_depth=255, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=790; total time= 0.7s [CV] END bootstrap=True, max_depth=255, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=790; total time= 0.7s [CV] END bootstrap=True, max_depth=255, max_features=sqrt, min_samples_leaf=20, min_samples_split=10, n_estimators=790; total time= 0.7s [CV] END bootstrap=True, max_depth=240, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=676; total time= 0.7s [CV] END bootstrap=True, max_depth=240, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=676; total time= 0.7s [CV] END bootstrap=True, max_depth=240, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=676; total time= 0.8s [CV] END bootstrap=True, max_depth=240, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=676; total time= 0.7s [CV] END bootstrap=True, max_depth=240, max_features=auto, min_samples_leaf=2, min_samples_split=10, n_estimators=676; total time= 0.7s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=664; total time= 0.6s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=664; total time= 0.8s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=664; total time= 0.7s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=664; total time= 0.7s [CV] END bootstrap=False, max_depth=75, max_features=sqrt, min_samples_leaf=10, min_samples_split=10, n_estimators=664; total time= 0.7s
RandomizedSearchCV(cv=5, estimator=RandomForestClassifier(n_jobs=-1),
n_iter=1000,
param_distributions={'bootstrap': [True, False],
'max_depth': array([ 10, 15, 20, 25, 30, 35, 40, 45, 50, 55, 60, 65, 70,
75, 80, 85, 90, 95, 100, 105, 110, 115, 120, 125, 130, 135,
140, 145, 150, 155, 160, 165, 170, 175, 180, 185, 190, 195, 200,
205, 210, 215, 220, 225, 230, 235, 240, 245, 250, 255]),
'max_features': ['auto...
736, 739, 742, 745, 748, 751, 754, 757, 760, 763, 766,
769, 772, 775, 778, 781, 784, 787, 790, 793, 796, 799,
802, 805, 808, 811, 814, 817, 820, 823, 826, 829, 832,
835, 838, 841, 844, 847, 850, 853, 856, 859, 862, 865,
868, 871, 874, 877, 880, 883, 886, 889, 892, 895, 898,
901, 904, 907, 910, 913, 916, 919, 922, 925, 928, 931,
934, 937, 940, 943, 946, 949, 952, 955, 958, 961, 964,
967, 970, 973, 976, 979, 982, 985, 988, 991, 994, 997,
1000])},
verbose=2)
Hyperparameter tuning can be very time consuming with RandomForrestClassifier, due to the number of hyperparameters tuned. Moreover, the number of Decision Trees has a direct impact on the runtime. Because of time constraint, the classifier tuned with RandomizedSearch will be used
#rf_clf=RandomForestClassifier(n_jobs=-1)
#param_grid_rf = {'n_estimators': [random_rf.best_params_['n_estimators']-5,
# random_rf.best_params_['n_estimators'],
# random_rf.best_params_['n_estimators']+5
# ],
# 'max_features': ['sqrt','auto'],
# 'max_depth':[random_rf.best_params_['max_depth']-5,
# random_rf.best_params_['max_depth'],
# random_rf.best_params_['max_depth']+5],
# 'min_samples_split': [random_rf.best_params_['min_samples_split']-1,
# random_rf.best_params_['min_samples_split'],
# random_rf.best_params_['min_samples_split']+1],
# 'min_samples_leaf': [random_rf.best_params_['min_samples_leaf']-1,
# random_rf.best_params_['min_samples_leaf'],
# random_rf.best_params_['min_samples_leaf']+1],
# 'bootstrap': [True,False]
# }
#rf_bp=GridSearchCV(rf_clf,param_grid=param_grid_rf,cv=20)
#rf_bp.fit(X_train,y_train)
rf_cv,rf_score,rf_acc_v,rf_acc_t=clf_score(random_rf.best_estimator_,X_val,y_val,X_test,y_test,cv=20)
Roc_AUC validation set: 0.9161666666666667 ROC_AUC test set: 0.8849512069851053 Accuracy validation set:0.9015544041450777 Accuracy test set:0.8695652173913043 Confusion Matrix: [[ 93 30] [ 6 147]]
rf_famd_acc,rf_pca_acc,rf_kpca_acc=plot_perf_comp('Random Forest Classifier',random_rf.best_estimator_,rf_cv.mean(),rf_score,rf_acc_v,rf_acc_t)
transformation_comp('Random Forest Classifier',random_rf.best_estimator_)
StandardScaler() Roc_AUC validation set: 0.9203333333333333 ROC_AUC test set: 0.8753354203935599 Accuracy validation set:0.8497409326424871 Accuracy test set:0.8659420289855072 Confusion Matrix: [[ 95 28] [ 9 144]] MinMaxScaler() Roc_AUC validation set: 0.9136666666666666 ROC_AUC test set: 0.8717387548234753 Accuracy validation set:0.844559585492228 Accuracy test set:0.855072463768116 Confusion Matrix: [[ 90 33] [ 7 146]] RobustScaler() Roc_AUC validation set: 0.9203333333333333 ROC_AUC test set: 0.8425058548009368 Accuracy validation set:0.8601036269430051 Accuracy test set:0.8442028985507246 Confusion Matrix: [[101 22] [ 21 132]] PowerTransformer()
C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\preprocessing\_data.py:3253: RuntimeWarning: divide by zero encountered in log loglike = -n_samples / 2 * np.log(x_trans.var())
Roc_AUC validation set: 0.9133333333333333 ROC_AUC test set: 0.8802401930523598 Accuracy validation set:0.8497409326424871 Accuracy test set:0.8695652173913043 Confusion Matrix: [[ 95 28] [ 8 145]]
ADABoost base estimator supports only estimators with the parameter sample_weight:
import inspect
from sklearn.utils import all_estimators
for name, clf in all_estimators(type_filter='classifier'):
if 'sample_weight' in inspect.getfullargspec(clf.fit)[0]:
print(name)
AdaBoostClassifier BaggingClassifier BernoulliNB CalibratedClassifierCV CategoricalNB ComplementNB DecisionTreeClassifier DummyClassifier ExtraTreeClassifier ExtraTreesClassifier GaussianNB GradientBoostingClassifier HistGradientBoostingClassifier LinearSVC LogisticRegression LogisticRegressionCV MultiOutputClassifier MultinomialNB NuSVC Perceptron RandomForestClassifier RidgeClassifier RidgeClassifierCV SGDClassifier SVC StackingClassifier VotingClassifier
from sklearn.ensemble import AdaBoostClassifier
from sklearn.tree import DecisionTreeClassifier
ada_clf=AdaBoostClassifier()
param_grid_ada = {
'base_estimator':[log_bp.best_estimator_,svm_bp.best_estimator_,
random_rf.best_estimator_,DecisionTreeClassifier()],
'n_estimators':[30,50,70],
'learning_rate':[0.02,0.04,0.06]
}
ada_bp=GridSearchCV(ada_clf,param_grid=param_grid_ada,cv=5)
ada_bp.fit(X_train,y_train)
print(ada_bp.best_params_)
ada_cv,ada_score,ada_acc_v,ada_acc_t=clf_score(ada_bp.best_estimator_,X_val,y_val,X_test,y_test,cv=5)
C:\Users\valentf\Anaconda3\lib\site-packages\joblib\externals\loky\process_executor.py:702: UserWarning: A worker stopped while some jobs were given to the executor. This can be caused by a too short worker timeout or by a memory leak. warnings.warn(
{'base_estimator': RandomForestClassifier(max_depth=55, max_features='sqrt', min_samples_leaf=4,
min_samples_split=5, n_estimators=607, n_jobs=-1), 'learning_rate': 0.04, 'n_estimators': 30}
Roc_AUC validation set: 0.9056871233341821
ROC_AUC test set: 0.8753354203935599
Accuracy validation set:0.8601036269430051
Accuracy test set:0.8659420289855072
Confusion Matrix:
[[ 95 28]
[ 9 144]]
ada_famd_acc,ada_pca_acc,ada_kpca_acc=plot_perf_comp('ADABoost',ada_bp.best_estimator_,ada_cv.mean(),ada_score,ada_acc_v,ada_acc_t)
transformation_comp('ADABoost',ada_bp.best_estimator_)
StandardScaler() Roc_AUC validation set: 0.9132499999999999 ROC_AUC test set: 0.8724956507099164 Accuracy validation set:0.8601036269430051 Accuracy test set:0.8623188405797102 Confusion Matrix: [[ 94 29] [ 9 144]] MinMaxScaler() Roc_AUC validation set: 0.9086666666666666 ROC_AUC test set: 0.8824893917963225 Accuracy validation set:0.8549222797927462 Accuracy test set:0.8695652173913043 Confusion Matrix: [[ 94 29] [ 7 146]] RobustScaler() Roc_AUC validation set: 0.9111666666666667 ROC_AUC test set: 0.8503205128205128 Accuracy validation set:0.8601036269430051 Accuracy test set:0.8514492753623188 Confusion Matrix: [[101 22] [ 19 134]] PowerTransformer()
C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\preprocessing\_data.py:3253: RuntimeWarning: divide by zero encountered in log loglike = -n_samples / 2 * np.log(x_trans.var())
Roc_AUC validation set: 0.9083333333333332 ROC_AUC test set: 0.8753354203935599 Accuracy validation set:0.8601036269430051 Accuracy test set:0.8659420289855072 Confusion Matrix: [[ 95 28] [ 9 144]]
Similar to ADABoost, voting classifier can only use estimator having predict_proba method.
from sklearn.utils import all_estimators
estimators = all_estimators()
for name, class_ in estimators:
if hasattr(class_, 'predict_proba'):
print(name)
AdaBoostClassifier BaggingClassifier BayesianGaussianMixture BernoulliNB CalibratedClassifierCV CategoricalNB ClassifierChain ComplementNB DecisionTreeClassifier DummyClassifier ExtraTreeClassifier ExtraTreesClassifier GaussianMixture GaussianNB GaussianProcessClassifier GradientBoostingClassifier GridSearchCV HalvingGridSearchCV HalvingRandomSearchCV HistGradientBoostingClassifier KNeighborsClassifier LabelPropagation LabelSpreading LinearDiscriminantAnalysis LogisticRegression LogisticRegressionCV MLPClassifier MultiOutputClassifier MultinomialNB NuSVC OneVsRestClassifier Pipeline QuadraticDiscriminantAnalysis RFE RFECV RadiusNeighborsClassifier RandomForestClassifier RandomizedSearchCV SGDClassifier SVC SelfTrainingClassifier StackingClassifier VotingClassifier
from sklearn.ensemble import VotingClassifier
estimators=[('log',log_bp.best_estimator_),('ada',ada_bp.best_estimator_),('knn',knn_clf),
('rf',random_rf.best_estimator_),('svm',svm_bp.best_estimator_)]
ens_clf=VotingClassifier(estimators=estimators,n_jobs=-1,voting='soft')
param_grid_ens={
'weights':[[1,1,1,1,1],[1.4,1.15,1,0.7,0.7],[1.4,1.2,1,0.8,0.6]]
}
ens_bp=GridSearchCV(ens_clf,param_grid=param_grid_ens,cv=5)
ens_bp.fit(X_train,y_train)
print(ens_bp.best_params_)
ens_cv,ens_score,ens_acc_v,ens_acc_t=clf_score(ens_bp.best_estimator_,X_val,y_val,X_test,y_test,cv=5)
{'weights': [1, 1, 1, 1, 1]}
Roc_AUC validation set: 0.908525026172085
ROC_AUC test set: 0.8640873015873016
Accuracy validation set:0.8497409326424871
Accuracy test set:0.8586956521739131
Confusion Matrix:
[[ 96 27]
[ 12 141]]
ens_famd_acc,ens_pca_acc,ens_kpca_acc=plot_perf_comp('Voting Classifier',ens_bp.best_estimator_,ens_cv.mean(),ens_score,ens_acc_v,ens_acc_t)
transformation_comp('Voting Classifier',ens_bp.best_estimator_)
StandardScaler() Roc_AUC validation set: 0.9061666666666666 ROC_AUC test set: 0.8640873015873016 Accuracy validation set:0.8497409326424871 Accuracy test set:0.8586956521739131 Confusion Matrix: [[ 96 27] [ 12 141]] MinMaxScaler() Roc_AUC validation set: 0.90575 ROC_AUC test set: 0.8746817538896748 Accuracy validation set:0.8549222797927462 Accuracy test set:0.8623188405797102 Confusion Matrix: [[ 93 30] [ 8 145]] RobustScaler() Roc_AUC validation set: 0.9263333333333332 ROC_AUC test set: 0.8434873949579832 Accuracy validation set:0.8497409326424871 Accuracy test set:0.8442028985507246 Confusion Matrix: [[108 15] [ 28 125]] PowerTransformer()
C:\Users\valentf\Anaconda3\lib\site-packages\sklearn\preprocessing\_data.py:3253: RuntimeWarning: divide by zero encountered in log loglike = -n_samples / 2 * np.log(x_trans.var())
Roc_AUC validation set: 0.9057499999999999 ROC_AUC test set: 0.8671098170631214 Accuracy validation set:0.844559585492228 Accuracy test set:0.8623188405797102 Confusion Matrix: [[ 97 26] [ 12 141]]
x_labels=['Logistic Regression','Support Vector Machine','K-Nearest Neighbors','Naïve Bayes',
'Random Forrest','AdaBoost','Voting Classifier']
acc_list_ori=[log_acc_testscore,svm_acc_t,knn_acc_t,nb_acc_t,rf_acc_t,ada_acc_t,ens_acc_t]
acc_list_pca=[max(log_pca_acc),max(svm_pca_acc),max(knn_pca_acc),max(nb_pca_acc),
max(rf_pca_acc),max(ada_pca_acc),max(ens_pca_acc)]
acc_list_famd=[max(log_famd_acc),max(svm_famd_acc),max(knn_famd_acc),max(nb_famd_acc),
max(rf_famd_acc),max(ada_famd_acc),max(ens_famd_acc)]
acc_list_kpca=[max(log_kpca_acc),max(svm_kpca_acc),max(knn_kpca_acc),max(nb_kpca_acc),
max(rf_kpca_acc),max(ada_kpca_acc),max(ens_kpca_acc)]
log_famd_acc,log_pca_acc,log_kpca_acc
def plot_class(x_labels,test_acc,name):
fig, ax = plt.subplots(figsize=(13, 10))
ax.barh(x_labels, test_acc, align='center', alpha=0.5, ecolor='black', capsize=10)
ax.set_ylabel('Classifiers')
ax.set_xlabel('Accuracy score of the prediction')
ax.set_title('Accuracy score of prediction of different classifiers using '+ name+' data')
ax.yaxis.grid(True)
plt.xlim([0.8,0.95])
plot_class(x_labels,acc_list_ori,'original')
Ensemble methods are very similar to the 'simple methods' in term of accuracy on the original dataset. However ensemble methods with FAMD features usually leads to a better accuracy (above 0.9 expected).
plot_class(x_labels,acc_list_pca,'PCA')
plot_class(x_labels,acc_list_kpca,'KernelPCA')
plot_class(x_labels,acc_list_famd,'famd')
This project could yield better results with the following additions:
#stratifying leads better results - probably because some subsamples predictions would be full of 0 or 1